 We have a very interesting session lying ahead of us, so the macroeconomic effects of fiscal policies. We have two different papers, and I think without further ado, we can just kind of step in into the first presentation, which is about fiscal policy in a networked economy. And I think John Storm Beko is online and will present to us. Thanks a lot. John, you have 25 minutes for your presentation. All right. Hi, everyone. Thank you so much for having me. I'm sorry I'm not able to be there in person. I had a terrible last-minute passport issue, but thank you for still including me in the program. I'm very excited to present this work, which is joined with Joel Flynn and Christina Patterson. So over the last few years, we've seen policymakers in the US and in the EU adopt many different forms of fiscal policy. We've seen undirected transfers, targeted transfers, targeted spending, and it's important to understand which of these instruments policymakers should use and how they should target them. Now, in many simple economic models, such as models with representative agents, these policies all have very similar effects. And so the question of which to use and how to target isn't important, but we know that in reality, these policies are different from one another, not only because they have different partial equilibrium effects, looting on to say households with higher or lower MTCs, but also that their general equilibrium effects depend on how policies propagate through complicated networks of relationships between households, for example, input-output linkages and regional trade relationships. So what we wanna understand in this paper is how the rich network structure of the economy shapes the propagation of fiscal policy and therefore the optimal design. So we're gonna speak to that question in this paper using both theory and data. On the theoretical side, we're gonna develop a model of how heterogeneity affects the propagation of fiscal shocks. So this is a model that, on one hand, is gonna be very simple because it's going to have a simple story of what recessions are. It's gonna say, recessions come from prices being fixed in the short run and therefore markets clearing through labor rationing. But the model is also gonna be very rich in the sense that it's gonna have many different ways that households can be linked with firms and linked through one another. So we're gonna have input-output linkages, regional trade, heterogeneous employment relationships and heterogeneous consumption preferences. And within this setting, we're gonna provide sort of a novel decomposition that says how it is that these sort of network forms of heterogeneity matter for fiscal multipliers. Then we're gonna take this model to the data and we're gonna try to understand which of these forms of heterogeneity are actually important in practice. So we're gonna estimate our sort of network, Keynesian multiplier, and we're gonna sort of communicate one main result, which is that despite many forms of heterogeneity mattering for fiscal multipliers, in theory, only one dimension is really important for aggregate shock propagation in practice. In particular, a policymaker who wants to sort of have as expansionary an impact as possible can just target policies toward households with the highest NPCs as they would in much simpler models and not think about, for example, the fact that lower NPC households sometimes might direct their spending towards other very high NPC households. Let me spend just a moment on the literature. So in the macro literature nowadays, we see many forms of heterogeneity being introduced into models to address all sorts of different questions. What we see as our role here is introducing into our model many different forms of heterogeneity and trying to speak to the question of which forms of heterogeneity matter for answering which questions. So what we're gonna show in this setting is that some forms of heterogeneity, in particular variation in household NPCs are going to be very important for different aggregate responses to different shocks. And then other forms of heterogeneity, such as, for example, the direction of household consumption varying across different households are gonna be important for the distribution of impacts of fiscal policies, but not for their aggregate implications. More broadly, we see this work as fitting into a line of the literature that starts at least as far back as Miyazawa and continues through recent work, like Anderson et al., that studies the propagation of shocks in disaggregated economic accounts in which one can sort of track the flow of a dollar through many different forms of economic networks. Okay, so with that, let me jump into the model. This is a model of labor rationing in a recession. And it's a model which is, on one hand, very simple, and on the other hand, allows us to incorporate many different forms of economic heterogeneity. So what's simple about the model? Well, it's only gonna have two time periods, a classic Keynesian short run and a long run, and it's only going to have one labor factor. What's rich about the model is that we have many different sectors and many different households, potentially with different preferences. So in this model, in any period, there's gonna be many prices for the many different goods. There's gonna be a wage in each period that we normalize to one. And one of our key economic assumptions is that we're gonna assume the real interest rate is rigid between the short run and the long run. You can think of that as reflecting, binding zero lower bound, sticky wages in the short run and sticky inflation expectations. So it's gonna be very important for the model that prices, the inter temporal price is sticky and therefore markets have to clear in a different way. What firms do in the model is gonna be totally standard. They're gonna be competitive and constant returns to scale and they're gonna produce their output using labor inputs as well as a vector of intermediate inputs that they source from all other firms. Households are going to be a little bit more interesting. Part of what they do is totally standard. They're going to choose their consumption and their labor supply in the long run period flexibly subject to budget and borrowing constraints. But we're going to assume that their first period of labor supply is instead determined by labor rationing, which I'll be more precise about in a moment. The government in this model is going to have sort of two different ways of intervening through policy. The first is that they can purchase goods. They do government purchases like infrastructure spending. So that means they can purchase different goods in different time periods. And then also they have targeted transfers. They can levy different taxes on different households. And the last and really key economic component of the model is that as I mentioned, household labor supply in the first period is not going to be determined by households labor supply curves. It's instead going to be determined by rationing, which we think is a reasonable model of labor supply during severe recessions. So what I mean by that is there's going to be some function, we call it a rationing function, that's going to map the vector of firms labor demands. So how much labor is demanded by each firm in the economy to a vector of how much labor is supplied by each household in the economy. So this is sort of a reduced form rationing function that we only impose one assumption on, which is that it satisfies labor market clearing. What firms demand in total is how much households supply in total. And this is what's going to allow markets to clear despite the fact that we have rigid prices. So markets are not going to clear on prices inter-temporally, instead they're going to clear on labor being rationed in the short. So that you can think of this as sort of a classic Keynesian model, but one with heterogeneity in the sense that there are many different firms linked to one another and there are many households linked to firms heterogeneously through this rationing function as well as through different consumption preferences. All right, so that's the model. Our theoretical results, I'll express how this sort of heterogeneity in the model affects the propagation of fiscal policy shocks. So we consider two different forms of shocks. The first is changes in taxes and transfers and the second is changes in government spending. And in order to understand how shocks propagate, we find it useful to distinguish between shocks general equilibrium impacts and their partial equilibrium impacts. So what I mean by a shock's partial equilibrium impact on first period final output is how much it changes first period final demand before households income suggests. So the simplest case to think through as just a government spending shock, the government spending shock might be directed towards certain sectors, not others. So this changing government spending is a vector which may be positive in some sectors where the government does more spending and zero and others. And that has some partial equilibrium effect on final demand in certain sectors. So this partial Y is a vector that says there's more demand for some sectors. Our first theoretical result expresses the vector of general equilibrium changes in output as a multiplier matrix times the partial equilibrium change in final demand. And you can think of this multiplier matrix as sort of a vector version of the classic Keynesian multiplier. You'll see it takes the form sort of of a one over one minus MPC. But what is MPC in our model is more complicated than in the simple representative agent model. And in particular MPC is sort of a matrix. So what is that matrix comprised of? It has sort of five different components. The first is it depends on the input output network that links how much more production is needed from every industry when there's more demand for a particular final output. Second, it depends on how much labor is demanded by firms in each of those different industries who end up producing more. Third, it depends on the Jacobian of this rationing matrix that says when each of those firms needs to produce more output, how much labor they demand from which households. Fifth, it depends on the MPCs of all those different households who are employed on the margin. And then fifth and finally, it depends on what are the goods that are consumed by those households when they do more marginal spending. So the way that we like to think about the sort of network Keynesian multiplier is through the following chain of events. There's some partial equilibrium shock like the government demands more infrastructure goods. And that means that there needs to be more production not only from infrastructure firms but also everyone upstream of them. That means that there's more employment of workers who work at all those firms and in turn, some of those workers are going to spend part of their income, meaning there's more marginal consumption and in particular, there's marginal consumption of some certain vector of goods. That consumption of a certain vector of goods is in turn an additional shock that enters the economy and propagates through that same chain. So it's the sort of classic Keynesian multiplier logic just now accounting for the fact that a shock might hit certain workers who spend on goods produced by other certain workers and so on. So it's really a matrix Keynesian multiplier. So that general result tells us about how the vector of economic outputs of different goods changes in response to fiscal policy. Another thing we might be interested in is collapsing that vector down to the change in total aggregate output. So that's what we're going to talk about on this slide, how heterogeneity affects the aggregate multiplier stemming from any fiscal shock. And in particular, we show that heterogeneity affects aggregate fiscal multipliers through three distinct channels on top of the sort of baseline Keynesian multiplier you might have in mind. There's three channels through which heterogeneity affects aggregate fiscal multipliers are the following. First, there's an incidence effect where some shocks may load on to households who have disproportionately larger NPCs. Second, there's a bias effect where shocks may load on to households who direct their spending towards other households with higher than average NPCs. And finally, there's what we call a homophily effect which is that there may be a correlation between the NPCs of households that shocks load on to directly and the NPCs of the households that those households spend on in the sort of second round of the multiplier. So we show in the paper that to third order in NPCs, these three channels exactly capture the deviations of the change in output in our economy, the deviations of change in output in our economy from a simple Keynesian multiplier. So that means the change in aggregate GDP resulting from some fiscal shock is equal to a Keynesian multiplier term plus three corrections. One for this incidence effect, one for the bias effect and one for the homophily effect. So in order to show you in a little bit more intuitive away what these three different deviations away from the Keynesian benchmark correspond to, let me now go through a series of examples that isolate them each. So in order to give this example, let's consider a two household economy where we have one household with high NPC, maybe it's half, another household with low NPC, maybe it's 0.1. And I'm gonna show you sort of four different possible cases for what's the incidence of the shock onto different households and what's sort of the spending to income network of how different households spend on goods produced by different other households. Let's start off in the case of a shock with uniform incidence onto the two different households and a sort of neutral network. So I'm showing in this picture, the dashed lines are the incidence of the shock. So we're imagining, for example, that each of these households receives a transfer from the government of 50 cents. And then the solid lines are how households direct their marginal spending. So in this picture, we're seeing that the shock has equal incidence under the two households and then households direct their marginal spending, half of it toward themselves and half of it toward the other household. So in this case, none of the three heterogeneity corrections are present and aggregate amplification is the same as you would have gotten in an economy with a single household with the population average NPC. Now let's consider an example that isolates the incidence effect. So here we have the initial shock all being directed toward the higher NPC household. So that means that sort of the first round of the Keynesian multiplier all happens at a higher NPC and then further rounds propagate at the average NPC in the population. That generates more amplification through that incidence effect. Next, let's go back to baseline incidence, which is uniform across households and consider a biased marginal spending network where both households direct their marginal dollars towards the goods produced by the high NPC household. Maybe that high NPC household is producing a luxury good or something. So in that case, we get higher amplification, not in the first round of the multiplier, which happens sort of at the average NPC in the population, but rather in all subsequent higher order rounds of the multiplier, which happens at the NPC of the highest NPC household. That's what we're calling the bias effect, the marginal spending being directed towards someone different on average than the average in the population. And then finally, let's consider the case where again, we have uniform incidents, but each household directs its marginal spending onto itself, i.e. onto other households with similar NPCs. That's what we were calling homophily. People spend on other similar people. That too generates higher amplification because the Keynesian multiplier one over one minus NPC is a convex function. So if you have sort of two segregated economies, one with low NPC and one with high NPC, that has more amplification than a sort of baseline economy in which spending is spread between the two. Okay, so that's all I wanna say about the theory. And next we're gonna try to map these channels to the data and try to understand, is it important that, for example, in practice, there are some states or countries where folks have lower NPCs and there's a lot of spending internal to those states that might generate something like homophily in practice. All right, so now I'm gonna tell you how we match this model to the data. That's something that we're gonna do in the US in the year 2012. What I was calling sectors in the model, in the data is gonna correspond to about 2,600 state and three-digit NAICS industry cells. And what I was calling households in the model is gonna now correspond to about 4,000 different state times income quintile, times age quartile, times gender, times race groups, plus two additional groups, capitalists and foreigners, which are gonna be receptacles for income flowing to those groups, which you can just think of as sort of low NPC groups. So our first theoretical result gave this sort of network Keynesian multiplier. They told us we needed to think about how when there was a partial equilibrium change in demand, the initial change in demand affected the total change in demand facing all firms through the input-output network. How the total change in demand for firms affected labor demand for workers through labor rationing. And then how the total change in labor demand for workers affected consumption demand through sort of directed NPC matrix. So we need to estimate each of those three sufficient statistics in order to take this model to the data. So let me just tell you sort of at a very high level how we estimate each of those three matrices. So first we need to match, we need to estimate this regional input-output matrix that says when there's more final demand for any firm in any one state, how much more demand is there for any other sector in any other state of the economy. So we do this using BEA data on the make and use tables. We assume that the input-output structure of the economy is the same in all states, but then allow for the fact that firms in say auto manufacturing sector in California may tend to buy their steel from another state, Massachusetts. And we can figure out that interstate trade pattern using commodity flow survey data. Okay, second, we need to figure out this rationing matrix that says when there's more output in any industry, how much more employment is there in any demographic group in the same state as that industry. So to figure that out, we use labor shares at the state and sector level, as well as information on the demographics of workers in each state and sector cell. So a simple sort of baseline assumption would have been to assume that when firms demand more labor, they demand it from demographic groups, locally proportional to those groups employment in their state sector. It turns out that though that there is another pattern in the data documented by my co-author Christina's work, which is that workers who tend to have higher NPCs are rationed more sensitively on the margin. And so we also account for that sort of high NPC bias, rather than just assigning proportionally to demographic groups employment in the state industry cell. Finally, we have to calibrate this directed NPC matrix that says when any of the households in our model get more income, on what sectors, on what goods are they spending their marginal income? So this has two components. The first is that we estimate PS, excuse me, we use the PSIT to estimate NPCs for each of the 80 or so demographic groups in the model. We assume those NPCs are constant across states. And then to figure out the direction of household NPCs, we assume that households with marginal spending has the same direction as their average spending. That's an assumption, but one that to the extent we can test it in the data appears to be reasonable. And then finally to figure out how much of the spending of a household in Massachusetts is on computers from Michigan versus California, we again use that commodity flow survey data on interstate trade. Okay, so now that we've calibrated all these aspects of this network Keynesian multiplier, we can go to the data and talk about what the multipliers look like in the calibrated model. So the first thing I wanna look at is just how much variation is there in fiscal multipliers? So that's what I'm showing you in the plot on the left, which shows just the range of fiscal multipliers for government purchase shocks that are directed towards different state industry pairs. Two things are apparent in this figure. The first is that there's sort of an average, our aggregate government purchases multiplier around 1.3, which is sort of in line with fiscal multiplier estimates in the data. But it's also showing that there's a lot of heterogeneity in this fiscal multiplier depending on the state industry pair that's targeted. So we see fiscal multipliers ranging from about 1.1 to 1.6 depending on which state and which industry is targeted. If we look instead at fiscal multipliers and transfer shocks, we see even more variation. Intuitively, that's because transfer shocks more directly target households with different NPCs, whereas government purchase multipliers sort of are spread across many different households employed in different sectors. So, so far this range of variation in multipliers is significant, but I haven't told you yet whether it's due to sort of just incidents of different shocks onto folks with different NPCs or whether it really depends on those rich network interconnections like the bias and homophily effects that we talked about in the theory. So there's one thing I want you to take away from this talk. It's that in practice, in terms of thinking about aggregate fiscal stimulus, these bias and homophily effects are really not important. So I'm gonna show you that through a figure and what this figure shows is on the X axis, the NPC of all the different demographic groups in our model and on the Y axis, the average NPC of the households employed in producing the basket of goods consumed by the households on the X axis. So there should be sort of two things that come out of this graph. The first is that the sort of basket weighted average NPC is very similar across the different groups on the X axis. So in that tells us there's not much scope for the homophily effect for households who are targeted by some fiscal shock to spend on other households who have very different consumption from very different NPCs from their own. Second, the level of basket weighted NPCs that's common across all households is very close to this dashed line, which is the income weighted NPC in the population. And that means there's not much scope for that bias effect that says households marginal spending could be directed to something very different from the average household in the population. So putting those two observations together, it means that in the data, there's just not much scope for bias or homophily, which means that the aggregate effect of any shock is just given by its incidence onto either high or low NPC households. Now, what's that coming from in the data because we expected we might have had some positive homophily, for example, from the fact that many households spend locally and NPCs vary across different groups. It turns out that in the data, there's both this sort of local spending effect. But there's also a countervailing effect, which is that higher NPC households tend to consume goods produced by low labor share industries, i.e. industries where a lot of income flows to capitalists who are low NPC households. So high NPC households spending on low NPC households, that's a kind of anti-homophily. So what we're seeing is that there's both a kind of anti-homophily from this labor share spending effect as well as positive homophily from local spending. These two effects are both relatively small and also counteract one another. So far, I've just been talking about aggregate effects, but the model also has distributional implications. I don't have time to talk about this much, but let me just say, we think it's very interesting. We can sort of tell you, in the calibrated model, how much of the stimulus coming from any fiscal shock happens within state versus how much spills over onto neighboring states. In the model, it's about a half. All right, and just about out of time. So let me wrap up. What we've done in this paper is sort of built a symbol, but at the same time, rich model to answer these questions about how different forms of heterogeneity matter for the design of fiscal policy. And if there's two takeaways I want to leave with you, it's that on one hand, targeting fiscal policy is very important because fiscal multiplier is very widely depending on how a fiscal shock is targeted. But at the same time, targeting fiscal policy is actually simpler than we might have thought because heterogeneity and aggregate fiscal multipliers depend only on the NPCs that those shocks load onto and not on these sort of higher order network effects as if the model of the economy was sort of simpler than it really is in reality. Okay, so thank you so much for listening. I'd be very happy to hear the discussion and take questions. Thank you. Thank you very much, Joe. So I give the floor now to Andrea who will give the discussion for your paper. Thank you. Thank you very much. Thanks a lot to the organizers for inviting me to discuss this very interesting paper. So the question in the paper I think is very clear. It's already been at the center of the discussion in yesterday's session. And it's a very classical question on fiscal policy, which is about the size of fiscal multipliers. Again, this is a classical question has been quite a bit of interest on it since the 2008 financial crisis. And again, with COVID, where we've seen various type of fiscal stimulus, general transfer, targeted transfer, targeted spending and so on. And so what the authors bring to the table to address this question is to study the role of heterogeneity and network linkages in this context and asking how these elements affect the conclusions from what we know already and what's in the literature. So it's a very rich and very ambitious paper. It has basically three parts, a very general, the authors call it semi-structural model with nominal rigidities, heterogeneity in the household sector and the network or input output structure. They make a very ample use of micro data for the calibration to really give a sense of the quantitative implications of the model, both on the overall size of the multiplier, but also on the distribution, which is, I think, very interesting and useful. And finally, there's quite a bit of welfare analysis through a number of counterfactual. Joe only had the time to discuss a little bit of that, but the paper goes into a number of additional experiments. In particular, I found very interesting what he alluded to in terms of geographical spillovers and so on. So the main finding, as has been already pointed out in the presentation, is that targeting households with high MPC maximizes the agri-multiplier. So it's a very stark finding in a model that is extremely rich and potentially has other implications, but I think the finding is extremely clear. There is a little bit of a role for IO linkages in compressing the distribution of the multiplier. This is a picture which is in the paper that was not shown in the presentation, but it's quite interesting to see how IO linkages kind of tilt the distribution of multiplier and I'll get back to that in a little bit. And as I said, I found particularly interesting the results on regional spillover. So for $1 of government spending directed to a certain state, the agri-multiplier is about 1.3, so that's kind of the average across the states, but there is almost a 50% effect that is due to out-of-state spillovers. So I think for people interested in this regional spillovers, both in the Euro era perhaps and more broadly in international context, I think this is an extremely interesting result. Okay, so here is just a picture that I found extremely, I guess, illuminating in terms of the main result, which is about targeting MPCs. So what you have on the left is basically a picture that shows on the horizontal axis the average MPC on demographic regions and on the vertical axis, the multiplier. You see that the relationship is basically it's very clear, very stark, it's linear. On the right-hand side, instead, you have the purchase multiplier. So the left-hand side is the transfer multiplier, the right-hand side is the purchase multiplier against plotted against the MPC and you see that the relation is not as clear. So here the intuition I think is that, if you give a dollar, I think it's a point that Olivier made yesterday. If you give a dollar to someone who's really constrained, so that the effect is quite strong, if you do it through government purchases, the effect is not so direct. And so that's why there is a positive correlation but it's not so clear. The other picture I want just to return to this, I really love this maps, is the extent of the regional spillovers. I think John showed the case of Michigan in the paper, they also have another experiment where they talk about this other experiment which is Texas, which you see here on the left. And so the purchases are targeted to a certain region but then the maps gives you the extent of the spillovers and you see that there is sort of a gravity effect, meaning that the states, the areas that are neighboring the state where the purchases occur are affected more strongly. Okay, so these are my main takeaways in terms of the results. My comments are gonna be on each part of the paper. And I don't have major issues, I guess, with the paper I think is very interesting. Part of it I think is a bit more transparency. Part of it I'm gonna maybe bring a little bit of other contributions from the literature just to make a couple of points. So in terms of the model, again, I guess the main issue that I have is this language, is the use of language that the authors use. I mean, they talk about partial equilibrium versus general equilibrium to me is very much more impact versus propagation. This is a model where prices, wages, and interest rates are always fixed. And I think that to some extent I am completely on board with the idea that this is a model of a recession where the economy enters a liquidity trap and so on, completely fine with the downward wage rigidities that are used. But I think especially when we start talking about government spending that targets certain sectors, thinking about prices not moving becomes a little bit extreme. The other point is sort of about the liquidity trap and the way I think they think about the liquidity trap in the sense that at some point there is a big shock, the economy enters a recession, nominal interest rate goes to zero. That's kind of the way I think about it, it's a hard constraint on policy. But then we've seen interest rates staying at zero well beyond the recession. And that to me is much more a policy decision, right? There's formal guidance, there are other reasons why central banks wants to stay at zero. And I guess the question is whether that matters for the fiscal stimulus. So I think it would be interesting to kind of relax a little bit the strong assumption and think about these different scenarios. The second point is that these multipliers that they calculate to me are really static, they're really impact multipliers. And when we get to compare it with the numbers in the literature, I think the literature, if my reading is correct, it's emphasized a lot dynamic multipliers. And so I'm not sure exactly how to read this comparison, whether it makes sense or whether there is something missing there. I guess more generally from perspective of the model, it wasn't entirely clear to me how these two periods matter. And sort of relatedly, I think it would be interesting to think a little bit about or be a bit more explicit about the source of marketing completeness that are necessary for the results. Lastly, I was a little bit surprised by the lack of the discussion on the type of fiscal rules. And I guess, again, this is somewhat related to the fact that this model is only these two periods, that they talk about balanced budget over two periods. I mean, that to me just means that the intertemporal government budget constraint is satisfied, but presumably depending on how you finance the fiscal expansion, so that's gonna have an effect. My understanding is everything is debt financed here, but I think a little bit more transparent than that would be useful. Okay, so the second set of comments relates to some of the results. And here, I guess because the results is so strong on the idea of targeting NPCs, what I wanted to do is to bring to the table a bunch of results from other papers that have focused a little bit more on the network structure, which is central in the model, but then it's not so central for the result except for the distribution and so on. So one paper by Bucaes, Raquedi and Santoro, which came out this year in the AJ macro, discusses how the network structure actually provides amplification away from the zero lower bound and dampening at the zero lower bound. I think that dampening at the zero lower bound is consistent with what's in John Joel and Christina's paper. And the intuition here is that basically in the network structure is a source of real rigidity, which makes inflation less responsive to shocks and so makes the need of the central bank to react to the spending not as strong, but then conversely, when you are at the zero lower bound, you have the opposite effect. So I think this is consistent with this paper. I think it highlights the role of having perhaps a more flexible structure on the pricing side. The second point, another paper by the same authors, which is a working paper, looks at targeted sectoral government spending, which is also something that this paper does. And lists a number of characteristics that are important for the aggregate response when the government target purchases in certain sectors. And so these characteristics are relatively small contribution to private final demand, low markups, high labor intensity, and downstream in supply chain. Finally, paper by Peri Rakedi and Varotto, which is coming out in a restart, that composes the role of government spending in public investments and government consumption, showing that the multiplier of public investment is actually larger than that of government consumption. So that's also something quite complimentary to this because it's done also in a network economy. Lastly, how am I doing on time? Okay. Right, so that's kind of the last comment. So lastly about the policy message. So I guess if you read this paper and you think about it normatively, it seems that the authors would suggest, maybe John can tell us more about that. The authors are suggesting, well, relying on transfers seems to be a good idea. In a sense, what they're saying is that expenditure is not as effective because at the end of the day, it's really hard to figure out how to do purchases so that ultimately houses with high NPC may benefit from that, whereas the transfer are very direct. Fine, I think that's a message that I buy and I guess maybe we've seen more of that also in practice. I suspect that governments will still do kind of expansion based on purchases. One message that I think comes out by paper by Gernert and co-authors is that empirically, if I read it correctly, there is quite a bit of concentration of federal purchases in sector with high levels of price rigidity. This is related to the paper I mentioned before by Bocchets and others, which suggests that by doing this purchasing in sectors with high price rigidity, there is less pressure on inflation and so there is a less aggressive monetary policy response and less crowding out. So that could be one criterion that we can think about in terms of the policy response. And lastly, and I guess again, relatedly in the paper by Bocchets and others that I mentioned before, they document how state and local spending differently from government spending is mostly oriented towards services which have higher rigidity. So that's another I guess message in terms of policy that could be complimentary to the one in this paper. So let me wrap up. I think it's a very interesting, very thought provoking paper. It contains an elegant analysis through the theory coupled with a very serious micro to macro calibration. So a very detailed use of micro data. The main finding is that targeting households with high MPC maximizes the aggregate multiplier, which is very clear, very stark. And finally, my three comments have concerned the type of model that they use, which is very general, but also relying on very specific assumption, bringing in something from the existing literature and then some insights about how government purchases are related to sectoral characteristics. Thank you very much. Thank you very much, Andrea. I would like to open the floor first for some questions here from the audience. And then I would give the floor back to you, John, so that you can respond both to Andrea's comments and also any questions that maybe the audience might have. So somebody's, yes, please. Hey, John. So I have a question on the connection between model and data in particular, how it relates to your main takeaway of kind of targeting by MPC. So kind of in your model, the short run, I guess once I take it to the data would correspond more to kind of business cycle frequencies. And I may have, you guys are kind of measuring impact MPCs or annual MPCs, but it may be that even low MPC households may have a non-trivial amount of spending, at least over business cycle frequencies overall, like not as concentrated at the beginning as the very liquidity constrained high MPC households may have, but still over like two, three years or something. And I think that, so these kind of cumulative intertemporal MPCs, those differences may be less pronounced than the impact MPC differences you're measuring. And that may lead to smaller differences in overall cumulative multipliers at the end of the day, right? Yeah, following up on one of the comments of the discussion, I wonder whether you could elaborate a bit on how period two matters for the model. So I understand there is an intertemporal budget constraint, but otherwise in the exposition of the results, I miss a bit the mechanics through which the closure works. Thank you. We have two more questions, yes, please. Yeah, maximizing the multiplier can be our objective maybe, but typically given that you have the model in the computer with Asians, you could just have... Sorry, I think you have to turn on the mic. Maybe you just, John, did you hear the question or? I heard the first half. Okay. It ended with given you have it in the computer. So given that you have a new computer, you can do some welfare analysis too, given that typically that's what we want to understand on policy, not simply who gets the highest multiplier, but he has more welfare effect. Well, then we have Ankel. So if we take your model at face value, does it follow that the best policy would be unconditional cash transfers targeted to the highest MPC households, valid only for period one and restricted to be spent in high marginal propensity to consume sectors? So what would be the implication in real life if you had to define the policy today? I'm wondering whether there's a trade-off between transfers and spending in terms of first and second moment, which is transfers seem to be much more reliable, but it's still the case that the MPC is less than one, where in the case of spending, the initial round is one. And I'm just wondering whether there is not some kind of indifference curve if you care about both moments, first and second. Thank you. I think I can give the floor back to you, John. All right, thank you so much for these excellent discussions and questions. I guess let me start by addressing Andrea's discussion. Andrea, thank you, that was very helpful. And I have to say, I actually agree with almost everything you said. And so I'll try to have sort of a complimentary response. One of the things that you emphasized, maybe the most was that our model is, in a sense, partial equilibrium in that while we allow for sort of one general equilibrium effect, namely this income multiplier, there's a lot of things we leave out, like price adjustments or heterogeneous stickiness and all these other channels may be important for some of our conclusions. So maybe it would be helpful if I could just say one of our conclusions that I think is sort of more robust and would still hold in a model with a lot of those other effects. So to sort of state what I think that claim is, let's consider a more general model that had all these sort of different effects in it, but still consider a shock that's sort of uniform in the sense that it's proportional to GDP. In that sort of model, I think our results would still imply that one of the channels you might have worried about, namely that this uniform shock would have more propagation due to sort of higher order multipliers happening at different MPCs than the first order multiplier. In fact, it's something you don't need to worry about. So why did I give the example in that way? It's because if you had sort of a uniform shock then we wouldn't need to worry about price rigidity's binding maybe in more sectors than others because there would be price pressure in all sectors. But you might have still been concerned that the sort of homophily channel, for example, might have caused much more amplification than you would have gotten in sort of a representative agent model. And so what I wanna say is just that this channel we're emphasizing about sort of network, income multipliers, is something that would show up in a lot of other models and our sort of main conclusion about the fact that that multiplier is sort of simple than you might have thought would carry over to some more general models even though those models would also have various other considerations within them. I also just wanted to thank you for following up on one of the things I didn't have a chance to mention which is this distinction between transfer and government purchase multipliers and the idea that in our model it's sort of better to be using government transfer multipliers because they target NPCs more directly. That's great, I think that's all I wanna say for now about the discussion but thank you so much, Andrea, that was incredibly helpful. So to the various questions from the audience, Christian, you mentioned that we use sort of one year NPCs but in fact in practice NPCs are dynamic so a household that doesn't spend money right away might spend it later in a recession. I totally agree with that and it would be great if we were able to do sort of a more inter-temporal version of our model. We have that in the theory in our paper actually but we don't have that sort of inter-temporal Keynesian process in our calibration. I think what that would do is it would sort of temper our results about how important is it to target high NPC households to the extent of course that you care about stimulus over the course of the business cycle and not just in the first year of a recession. I think our conclusion would still hold up though that you don't need to worry about these sort of higher order network effects. Those are just so small in the model that even if some heterogeneity and timing of spending turned them up a little bit I don't think it's likely to affect them sort of very much. The second question from the audience was about what's the role of period two in the model and sort of how does the model closure work? So I'm not sure whether I can answer this question satisfactorily but let me just sort of explain the dynamics of the model a little bit. So as I mentioned, the key friction in the model is that prices are sticky in the short run and that means that markets have to clear through some other mechanism which we assume is labor rationing in the first period. In particular that means that sort of the right way to think about what is total output in the first period is what is total spending in the first period and total spending can be different from households desired total labor supply. So the function of the second period is just so that households have somewhere to put their income that they decide not to spend. So there's some notion of an MPC that's different from one because households can decide to instead of spending today and therefore contributing to consumption and so output in the economy, they can save their money and leave it for the future. That's sort of all that the second period is doing in the model. Sorry, I'm trying to remember the next question. So I guess there were sort of several questions about what's the right policy? Is it really to just sort of maximize output as much as possible? Or maybe are there other reasons why we'd want to be doing some government spending? Something we have more in the paper but I didn't get to talk about today is that there are of course other motives for spending other than just maximizing output. There may be some direct value of government spending and that's something that the government may wanna do more of in a recession because there's a sense in which government spending is cheaper when workers are underemployed because the social cost of employment is less than the market wage that you might pay them. So indeed you might wanna do some amount of government spending. And sorry, now I'm remembering the other question in which is why would we wanna be maximizing output instead of thinking about some sort of welfare function and trying to redistribute income toward the workers who are underemployed? That's something we actually do in the paper though of course we have to make a lot of strong assumptions about who's underemployed. In the paper we have a section looking at the Great Recession and which commuting zones seem to have the most underemployment in the Great Recession. And then we ask from the perspective of the model under some assumptions with about how do wedges scale with underemployment? Where would you wanna be targeting stimulus? As you would expect the model spits out that you wanna target stimulus more or less just toward the places that have underemployment because that stimulates employment there and sort of addresses the labor wedges. But of course in the background in the model there's all sorts of complicated considerations about you might wanna spend on a certain demographic group in Hawaii because they turn out to spend a lot of money on another demographic group in California that's underemployed. Okay, so I'm sure I haven't answered everyone's questions fully but thank you so much. It's really thought provoking and I appreciate the feedback. Thank you very much, John. It's a shame that you're not here because I'm sure if you were here at the coffee break there would be some people who would come to you and ask you again some questions. And so thank you very much for being here and giving this presentation. Also thank you very much, Andrea for this very good discussion. And with this I would like to close this part of the session and then we would go to the next presentation. And I have here next to me needs wehrhofer from the Deutsche Bundesbank who will tell us something about the spending effects of fiscal transfers in a pandemic. Thank you. Yeah, hi everybody and thank you to the organizers for including our paper in the program. I think it fits really nicely into this session and it's gonna be a straight empirical paper but you can think of it as like a particular application of what we talked about already in the first paper. So when is the NPC gonna be high and what factors determine that? And the specific application here is during the COVID pandemic in Germany where there was also some fiscal transfers. And the paper is joined with two other colleagues from the research center. So the usual disclaimer applies here. Oh, I need to, perfect. All right. So I guess I don't need to motivate this a lot but fiscal stimulus is particularly important. We talked about this a bit already yesterday when monetary policy is constrained and it can depend quite a lot on the context. And I guess the COVID-19 recession was quite a different recession from all the other recessions that we think about beforehand and there's some particular features that I wanna highlight here. So we had a lot of supply side restrictions. This is not only because of supply chain issues but also literally because there is shops closed down there's restrictions on what you can buy. There's all kinds of supply side restrictions. And on the other hand, of course, and I'm gonna be talking here about the early phase of COVID. There was a substantial infection risk and people might change their behavior due to that. And all of these factors lead households to have high savings rates because they cannot spend on particular goods. So they might just save. And we wanna ask in this paper how do these factors impact the NPC of fiscal transfers? And the application that we're gonna be looking here is the German child bonus. So that's a direct payment to families in Germany who have children under the age of 18. And we were gonna be evaluating this using a scanner data from the gift card. So just a few quick words on the literature. So there's lots of work on the fiscal policy response to COVID-19. So I mean, here on this slide there's only a few papers mainly looking at the US stimulus payments. There's of course a huge literature looking at other recessions. But what is quite interesting is that they find quite different NPCs. So just looking at the same transfers in the US, you can find NPCs between 10% and 46% in the literature. And typically these papers find that, and this is connecting nicely to the paper beforehand that the NPCs hire for liquidity constrained households. Secondly, there's been a literature which tries to connect stabilization policies and the pandemic conditions and how these two interact. And there's been some theoretical work which basically says that there's a two-way feedback effect. So if we have more economic activity, people go out more, people interact more, there's sort of more infections. And at the same time, if there's more infections, people might be more cautious and they might not go out so much. And this then has a feedback effect on the sort of on the epidemic. And what we provide here in this paper is basically empirical quantification of these effects. And there's already been some effect on this, especially looking at stay-at-home orders and how they impact fiscal multipliers. So a few quick words on the policy that we're looking at here. So this is the German child bonus. So this is a direct transfer to families which is paid on top of the regular child benefit. And this happened quite early on in the pandemic. So this is September, October, 2020, and then May, 21. That's the transfers that we're looking at. And in total, this was like for the average family, a transfer of almost 700 euros. So it's not as large as the US transfers but still is a sizable transfer. And what is quite interesting and quite neat for us in terms of identification is that the date when the transfer was paid out depends on the child benefit number, which is basically a random number that you get assigned the first time you apply for the benefit. So it's just on an ongoing basis. You get a number between zero and nine, depending on when you register with the government. And this number is gonna determine when the payment is gonna be paid out. And we're gonna use that random variation for identification here in the paper. The transfer itself was paid by automatic transfer, a bank transfer just as the normal child benefit. And it was announced in the middle of 2020 as part of the overall stimulus package by the German government and received quite some media attention there. In terms of how targeted the transfer was, so I mean, I would say in general, it was not very well targeted. Basically, everybody gets it, but the top 10% of the income distribution has to pay it back when they do the tax declaration at the end of the year. So it was somewhat targeted at the bottom 90%, but not very well targeted in general. And even those people who sort of had to pay it back in the end, I guess there were quite some of them who didn't know about that and might still have spent the money. All right, on the data, so we're using scanner data from the GFK. You can think of this as the German equivalent as of the Nielsen scanner data in the US. So this is detailed information on the household level on every purchase that they make at daily frequency. So these households are part of a panel and they always scan all their purchases and we basically get the information of what they bought, how much they discussed, where did they buy this and lots of other information. It covers all non-durable goods and semi-durable goods. So when I say semi-durable goods here, I mean, cloth, shoes, small electrical appliances, stuff like that, there's not gonna be the large durable, it's like washing machines in there. The sample period that we can have a look at is we have 2019 data and then we have data from July 2020 until June 21. So mid-2020 to mid-21. And interestingly, we can also look a bit deeper at this data and distinguish between in-person spending and online shopping, which might be one way that households avoid the issues during a pandemic of being infected or sort of shops being closed down. In total, we have about 8,000 households in the sample and roughly 20% of them have eligible children, so they're eligible for the transfer. We then merge this data with some other information about the case incidents of COVID at the county level and sort of very precisely at the county day level. And then we also have very, very detailed information about restrictions, which is provided by the German government, which is a daily data set, which tells us whether or not schools were closed down, whether not shops were closed down, whether there were masked mandates, curfews, certain mass social distancing requirements. So we basically know the severity of restriction for the households at each point in time. So this slide basically explains our identification here. So what we were able to do is we were able to field a survey attached to the scanner data. And in this survey, we asked households for their child benefit number. We do this in a somewhat indirect way. So we asked them when did they receive the usual child benefit within a month? And this then allows us to have a one-to-one mapping into the receiving of the child bonus payment. So basically, once we know when did you get your usual transfer in January 21, we can map this back to when did you get your child bonus in September, October, and May. Because basically for every number from zero to nine, there's a fixed payment date, which is publicly available. And just to check whether or not this is actually a random number. So you can see here a simple regression where we just regressed this number on a bunch of observable characteristics. So demographics, how much they spent in the month before receiving the payment, their income, their wealth, some other personal characteristics, and it doesn't correlate with any of them. So it truly seems to be a random number. Then coming to our estimation, it's very, very simple. It's a difference in difference regression where we basically use the differential payment dates and compare households who have received the transfer versus those that do not or have not received the transfer yet. In some specifications, we're also gonna control for the ongoing pandemic dynamics at the daily level. And we're also gonna have county times day fixed effects. So basically we are only comparing households who are living in the same county on the same day, but one of them has received the transfer, the other one has not. And we do basically the same thing also in a dynamic fashion. So looking at the days before receiving the transfer, looking at the days afterwards and seeing whether or not our assumptions that we have to make here for this to be a causal estimate hold. So in particular, whether or not the households were on similar spending trends before receiving the transfer. And these kinds of specifications have gotten a lot of attention in the econometrics literature in the recent years due to potential heterogeneous treatment effects. So we check whether or not this plays a role in our setting here using one of the new estimators by Sun and Abram. But it's not gonna make a big difference for us. All right, so I think I can just jump into the results here and start with a baseline MPC for the first transfer payment in September, 2020. And you can see we have four specifications where we consecutively add further controls. So we control for COVID, we control for date times county fix effects. And then we even interact the COVID controls with whether or not you're a parent and all specifications, you can see that we get a significant MPC of about, well, 11 to 12%. So it's significant, but it's rather small. And this is a quite robust result. So we do find some spending effects, but they don't seem to be very large. And of course, one has to keep in mind here that we're not covering all goods. So there might be some effect on durables. There might be some effect on services, which we're not covering at the end of the presentation. I'm gonna give you like a rough estimate, like even if you were to account for these, how high would the MPC be? But this is our baseline estimate here. And here you can see the dynamic picture and you can see that before, before receiving the transfer in the days before, households are on similar spending trends. And then after receiving it, they diverge. But you can see also that it's a quite noisy pattern and it's not a sort of very clear, huge increase in spending that households show. So I'm not gonna bore you too much with all the robustness checks that we do. I'm just gonna mention here that if we, so for the baseline specification, we only looked at one month of spending. If we sort of roll this further in time and we look at three month windows, the MPC increases somewhat to 21%, but it gets noisier and noisier over time as these estimates tend to do. So we're not comfortable extrapolating further further than three months out, but still 21% is not a huge MPC here. What is quite interesting is that we can now look at some of the mechanisms that are emphasized in the theoretical models which try to look at the interaction between a pandemic and spending behavior. And one of those mechanisms is that when people go out and spend, they have more contacts. There's more economic activity going on. And in our scanner data, we can actually see where people go to buy and how many shops they visit a day. So we can actually count how many shops that you go to to buy your goods. And I mean, ex-ante, this is not clear. You could just do the normal regular number of purchases and just go to the shop, the number of times that you would have gone anyhow, but just buy more when you're there. But we actually find that people sort of increase the number of economic interactions. And thereby might contribute to the ongoing pandemic. We also do a bunch of placebo tests. So we look at 2019 data, for example, and don't find any effect there. And also we look at the announcement of the policies. And also there's no effect that we can find there. All right, so let's dig a bit deeper into the data. So the first split that we're gonna make is we're gonna distinguish between semi-durable items and non-durable items. And our results are completely driven by non-durable items. So that's mostly food items. And also if we sort of distinguish between in-person spending and online spending, you can see that the percentage effect for online spending is quite large, about 20%. But since online spending is a relatively small share of the overall consumption basket, the MPC that is implied by that is very small. It's like about 3%. So most of the MPC is coming from in-person spending, but there's gonna be some interesting heterogeneity there that I'm gonna show you in a bit. So what we do next is we wanna understand why is the MPC so low? And what we do is we basically do treatment and heterogeneity analysis by regional characteristics. So we look at the COVID case count at the daily level. And what we can see is that if that is below the sample median, the MPC is actually way higher. So all of our results are driven by people living in counties at a time when the case rate was quite low. And in case the case rate goes up, we basically find insignificant estimates. Interestingly, if we do the same split by COVID restrictions, so when the restriction index is high or low, there doesn't seem to be a huge difference. And we also look at more traditional things that you would think about when doing these splits. So for example, looking at regions with have a higher unemployment rate, higher incidence of short-time work, there we also don't find lots of heterogeneity. And then if we zoom in into this effect on the COVID incidence and we split again by in-person shopping and online shopping, you can find that if the COVID incidence is high, people don't do in-person shopping, but they switch somewhat to online shopping. So basically you have counteracting effects here that if the incidence is high, people switch from in-person shopping to online shopping. So there seems to be, households seem to adapt somewhat to the conditions, but as I was mentioning beforehand, since the share of online shopping is so low, in the aggregate, it doesn't make a huge difference. We also have information at the individual level, and here I guess this connects to the earlier paper. If we look at households, who if we ask them in the survey whether or not they are constrained, and they answer yes, this is the largest effect or the largest heterogeneity that we can find at the individual level. So those households have an MPC of about 25%, but as you can see from the confidence bands, we only have like seven or 8% of them in the sample. So it seems like most households were not constrained when this transfer was paid out. Another thing that I would like to emphasize here is the effect by the number of children, which is basically a proxy for the size of the transfer. The more children you had, the higher the transfer, and the MPCs for these groups, whether or not you had one child or more than one child is remarkably similar. Then next, I mean, there were two more payments, one in October 2020 and one in May 21, and I haven't been talking a lot about them and there's a reason for that, because if you look at the estimates, you basically find nothing. So flat zero MPC that we get for the October 2020 package and the same happens in May 21. So both of these payments, we cannot find any effect. And also if we look into heterogeneities, it's very hard to find something here. So if we then come back to the aggregate effect of this transfer payment and we take into account these two null effects for the later payments, the aggregated MPC, even if we go three months out, it's quite low, it's only about nine percent. And as I was mentioning beforehand, even if we were to assume that there were similar effects on durables and services, this would only push up the MPC to about 25% and this is, I would say already, it's quite generous assumptions that we're making here, given that we didn't find any effect on semi-durables. So having sort of a similar effect on durables is quite a stark assumption, I would say. So the question is what changed over time to push down the MPC? I mean, one difference which might be obvious is that the transfer size was different, but as I was mentioning beforehand, if we look at families with one child or more children who got differently sized transfers, there's no difference in the MPC. So it's unlikely that this explains the null effects. The macro conditions were also quite similar over time. So if we look at this month and we look at the unemployment rate or the short-time work rate, so you can see here in the dashed lines, the month when the transfers were paid out, they're almost the same over time for these three months. So this is also an unlikely explanation, but given our heterogeneity results for the COVID case rate, one obvious place to look for would be what the pandemic was doing in the meantime. And as you can see in the first transfer period, in September 2020, case rates were quite low because we were coming out of the summer and then they increased a lot in October and then basically stayed high for the winter and came down in May, but we're still quite high in May. So our best explanation basically for these transfers not doing much is that they were badly timed during times when people were not going out and spending, but they were rather staying home. And I guess one important thing to emphasize here because I mean, this only happened a couple of years ago, but it feels like forever. This was a time when there was no vaccines. This was a time when there was like the original variants. So it's not comparable to sort of nowadays, thankfully. So, but our best explanation basically here is that when the government paid out these transfers in September and people were still quite optimistic and they were not sort of concerned a lot about COVID, people did go out and spend, but they didn't in the later month when COVID was sort of all over the place. All right, so let me conclude here. So we look at the spending effects of the German child bonus and we do find some significant effects for the first transfer round and the most stark feature which explains the level of the NPC is whether or not there was substantial case rates of COVID going on at the time or not. And then the second and third round, we don't find any effect and this might be related to the higher infection numbers during this month. So overall, depending on the assumptions that we make, we find the NPC between nine and 25%. And I guess one lesson from our paper is that the stimulus was way less effective when case rates were high and the same stimulus could have been achieved at way lower cost if it was timed correctly, exposed, I mean, it's very hard to sort of expand to think about that and households in general seem to have saved a substantial amount of that transfer and I mean, this might have contributed to them and if you look at aggregate saving rates during the pandemic, they were super high, especially also in Germany, this might have contributed to them spending them at a way later date when the COVID pandemic was kind of already over. So thanks a lot. Thank you very much, Mius. And we have our discussion, discussing Aleka Bachar from the Norwegian Business School. Yes, thank you very much for inviting me and discuss this very interesting paper. I really enjoyed reading it and yes, so let me jump right in with some, yeah, to quickly summarize. So as Niels was explaining this paper studies, the spending effects of three child bonus payments with the focus maybe on the first one in Germany during the COVID pandemic and what the authors do, what I think is really cool is that they combine this daily scanner data with a special commission service. So they are able to kind of map the survey respondents to the daily scanner data and that allows them to identify the NPC out of these bonus payments and what they do to identify is that they exploit this random variation in payment states across a given month and the payment date depends on the last digit of the child's social security or benefit number, which is also quasi-random because that depends on when you apply for the first time. So basically for your first child for the regular child benefit. So even after the pandemic, you get like these regular child benefit payments every month. And what they find is that they find significant spending effects all by low out of the first payment. So yeah, depending on this specification, but let's say around an NPC around 12%, however, they do not find any effect of the last two payments. And then in the second step, they also kind of relate their findings to the overall macroeconomic or pandemic situation and to find that the spending effects are weaker when infection rates are higher. And that is also kind of irrespective of government restrictions. So it's really about the infection rate that prevents or makes people not spend the money or not go out and spend. And also what I think is also interesting or nice to think about is that they also really can show that this payment increased or the first payment, especially increased the number of shop visits so possibly contributing to the spread of COVID-19. So kind of pointing out this maybe unintended side effect or also adverse side effect of this payment, which I think, yeah, I mean, it's pandemic specific, but it's interesting to think about which might not, yeah, like policymakers might not have had in mind when they introduced this policy. Okay, so let me maybe jump directly to the comments. So here, this picture Niels also showed us. So these are the daily or the dynamic effects of the first payment in September, 2020. And what you see also as Niels was pointing out, you see the kind of the no effect prior to the spending and then you see a jump in the spending. However, so it's a significant after four days. However, they are kind of noisy or somewhat imprecise. So I wanted to start by making maybe two suggestions which I think maybe, I don't know, may help to increase the position or reduce measurement error in these effects. And the first is that you maybe can do a little bit more to elicit really the exact payment received. Because I mean, if you exploit these daily variation and spending, I guess it is very important to really figure out when exactly households receive these payment. Because right now, as I understand, you assume kind of a two day lag between the day the payment is made and then it's received on the bank account of the households because it's paid by a normal bank transfer. And I was first, I was thinking, okay, typically these transfers or SIPA transfers take one business day. So my first question would be just, okay, what happens if you reduce this lag to one day and then also whether or not you account for weekends. So for example, if the payment is stayed on Thursday, then two day lag would be Saturday, but then of course the payment is only arrives on Monday. So kind of just try a little bit more, or maybe you did it, but like try a little bit more to really elicit when people receive this payment. And then I'm also wondering, so I mean, I did a little survey among some friends whether these payment actually arrive on time or whether there is some, I mean, one or two day lag on when these payments arrived and they told me they don't arrive at the same day every month. And I think it might also be a reason why you kind of find the significant effects only with the four day lag because it could be just that it takes some, like or that there's just some noise when these payments arrive. And then the second comment I had is maybe you could split the sample or control for the age of the children, because okay, now you said they are eligible until the child is 18, but I think for the regular benefit, at least you are eligible until the child is 25, as long as the child is still in educational training. And then if the children are old enough, so parents, because I mean, it is a child benefit transfer, so parents might transfer the money or give it to the children to spend. So I was wondering what happens if you kind of, first of all, whether you observe in this scanner data, if you let's say have a child which is 16 and they have independent consumption, kind of whether you observe that in the data or it's really only like the head of the household or their spouse that spends. So whether, I don't know, if they buy like some clothes for themselves, do you observe that? Because I'm wondering, maybe you could also kind of increase the precision if you drop households who children have moved out, but are still eligible for the benefits which are just different households or just allowed old enough to spend their own money or whether, yeah, kind of to separate a bit more independent consumption of children and parents. So that was just two ideas, which I thought maybe that could help you to get more precise estimates. The second comment I have is kind of regarding the null results or that you don't find any effects of the last two payments, which you argue is driven by the higher COVID incidents, which I think makes a lot of sense. However, I would kind of not discard some other mechanism that could strive. So you are also alluding to the size of the transfer. So the last two payments were smaller than the first one. And you kind of rule out this channel a bit because you don't find any differential response across households in the first payment with a different number of children who get like different sizes of the transfer. However, I'm kind of struggling a little bit with that because I'm thinking, okay, these are, I mean, it is very different if you have one, two, or three children and like consumption needs. And if in fact the money is primarily spent on children, which I think is a bit the intention of these transfers, I would expect the same response maybe. So because if I kind of spent the same amount of on each of my children, then if I have two children, I also spend double the amount roughly, I mean. But just kind of, I'm a bit struggling with this justification that it's not the size of the transfer. Then I was also thinking, maybe it's also just kind of the timing from the last transfer. So in the sense you get this transfer the first time in September, 2020, you spend it on something. But then you get it again one month later, but then you already kind of have these, you already have this extra consumption in the month before and then you might like need additional extra consumption. So that is just like the first surprise effect or not surprise, but the first effect is just largest. And then one thing, so here's just also this graph you showed, so that is kind of your show that the incidence is higher in October and May. But I was just wondering because I mean, there has been back in 2020, there was a lot of these stockpiling going on that households then kind of in anticipation maybe, went out and bought a lot of pasta and toilet paper. So I was just thinking that like kind of in September, people were kind of feeling, okay, it's now picking up again. So whether there was also just this increased consumption in September was just stockpiling of basically the winter in a sense what they could be spending the money on. And yeah, so just maybe to further strengthen your channel on the COVID incidents or some questions to that, I was wondering whether you could kind of compare households and counties with the same incidents across payments. So like let's say high incidence counties with low incidence counties in September and October, but they're kind of just controlling for the number of infections whether you kind of see whether households reacted the same to strengthen this incidence channel. And then another question, if it is, so here what you see here on the left figure is the right, but on the left, you see the aggregate consumption in 2020 or the monthly spending by households in 2020. And if the weak response in October is driven by high infection rates, it's just maybe like a wider, like the aggregate spending increase in that time. So that kind of goes maybe a little bit against your channel, which I think it's just would be nice to just clarify a little bit. Okay, so one small comment on the anticipation of the payment. So I mean, these transfer and also the exact payment date were fully anticipated by households. So in the sense, the theory would predict very low, very low spending effects of this around the exact timing of the date of the payment. You do test for the announcement effects in June 2020. So when the policy wants announced and then you also don't find any effect, which is I think very nice. However, I'm just thinking that and maybe that households smooth or they have budget for a given months. And so therefore you really find small effects because you exploit this daily variation in a sense that, yeah, households, they might not react three months prior to that, but then given when the month starts or kind of, I don't know, they do anticipate this effect and they do know, okay, this month I have a couple of hundred euros more in my overall budget. And then you just don't find any variation in this daily pattern. Okay, and then my last comment is maybe it's more what I think some more kind of to elude a bit more the implication beyond these direct effectors I think would be what you could further do in this paper or maybe another paper because you focus now mostly on the one month or then also as you said, up to three months after the payment receipt. But I mean, it is quite a short timeframe. So I'm just thinking that it would also be interesting to study more the long-term effects of this payment. So for example, if households mostly save the money and kind of all spend it after the pandemic, did that contribute to inflation? Is it somehow possible to quantify the effect or I was also thinking, could you kind of relate it to other policy measures that were implemented or that could have been implemented? So was it better to spend like each euro on these type of extra transfers or would it have been more helpful for households with children to, for example, I don't know, increase the number of paid sick leave days for parents so they could kind of help children with school. So kind of these type of just different types of measures that might have helped families that I think would be very interesting to kind of put it more in a broader context on the policy measures during the pandemic. And then if for example, this September 2020 response was a bit stockpiling in a sense that whether that just triggered an inter-temporal shift in consumption overall, like even absent of the payment so that basically consumption October might or in the next months might be lower than usual because people stockpiled a lot. And so basically you find zero effects is actually kind of a positive effect because the baseline might have been lower. Yes, but these are just kind of what you, I think what you could kind of bring the paper a bit further or maybe another paper follow up. Let me skip that, but yes, so just to conclude. So I think it's a very interesting paper with a very neat identification strategy, especially the data collection effort of the data is very cool with these scanner data plus the survey, which I think is also quite unique. And it's important that it really highlights the importance of the macroeconomic context when we study MPC. So in this particular case, the pandemic context, which we should always keep in mind when we think about these transfer payments. And yeah, my main suggestion would be first, I think really this trying to figure out the exact day of the payment receipt. And then also, which I think would be just interesting to see more the long-term effects or the broader context of this policy. Yes, thank you. Thank you very much, Annika. Then I would like to open the floor first. Yes, we have already one. So let me pick on Annika's comment. I think why the MPC zero is as interesting as why it's positive, right? And one thing that I was thinking is like in the first transfer, you can think that more transfers are coming later in the future. So suddenly, that system didn't existed. Now it started, you say, as long as there is COVID, money is gonna come, it's gonna continue flowing in. And maybe the next transfer you think, well, maybe this is the last one, and then you keep it. You see, so there could be, so we know that MPCs depends on the structure of the shock and that would be important. And another thing just to compliment Annika's, is you can give the money to your children. Do you have the cash instructions? Because you don't have the card for the kids, right? But maybe you can figure it out if they took more cash for the kids to go out or buy ice cream. I don't know, but that's it. Thank you. In a way, it goes almost back to a question that I had because the first payment was in some states, I think we still had school holidays. The type of restriction might have also mattered, but first I would like to collect more questions from the floor. Any, yeah. So Annika made that point, but I want to emphasize it. I think it's important, the idea of anticipation of payment. So your control group or households in your control group know or can be expected to know that they will be receiving this payment soon within days. And so aren't you worried that you are, but in this situation, this different, different regression would tend to underestimate the MPC. And then relatedly, you know, these are non-trivial payments, but they aren't that big either. So the idea that people would within one or two days run to shops and spend the money, that may not be accurate. So if they wait a little bit longer, again, because of the setup of the experiment, that you no longer have a difference between the control and the treatment group. And so aren't you worried about that? That it would also tend to underestimate the MPC. Yeah, I would like to hear a little bit more about this two days delay, because while you were talking, I checked this, I never bothered about this date. And I realized that was quite over the last couple of months. I mean, I got this money on the 17th, then on the 14th and 13th of the month. So there seems to be a variation in these days. More than two days, I was not aware, but now I checked them. So what do you do about it? Yeah, venturing a bit outside the scope of your paper dropper and trying to make a link with the, perhaps in a bit of a bold way, with the previous paper. I mean, one of the general conclusion of paper that the program was not very effective as a stimulus. So if one had to try to come with implication of what would be more effective, I'm tempted to say it's something that along the line of what's suggested in a comment with the previous paper. So you should do something which is target, even leaving aside the effect of COVID. So we're speaking of a deep recession now. So we abstract from the COVID something, would you agree would be something targeted to much more targeted, I mean, to liquidity, to constrain household, probably perhaps concentrated on durable and with some kind of time effect. I mean, to avoid also that household save the money would be a similar prescription that you get. Yes, I give the floor back to you. Okay, perfect. So I'm going to try to start at the end and then work my way through it. So I would definitely agree that a more targeted measure would have been way better. I think the reason the German government did this is that this is technically speaking, the only way the German government can transfer money to people in Germany because it's the only direct transfer to, so the technical infrastructure was in place. So they just did it. I think that's the main reason they sort of did this very unfocused, they didn't have the connection to like taxable income, for example, to make it more focused. So I mean, that's another debate, but I think that's the reason why it was so untargeted. In terms of like focusing on more durable goods, I would definitely agree. And there's other measures that the German government took, for example, the temporary VAT cut for, which basically emphasized this, like, okay, the VAT is going to be lower for half a year, then it's going to be higher. So if you want to buy a washing machine, do it now. So I think that was a way better stimulus program that was implemented. In terms of the two-day lack, so yeah, I mean, that's something we also talked to the Vermeer and Kasse, so the people who are basically doing this, and they told us that sort of two days is the minimum that we should assume, and then there might be some people who get it a bit later. And this is then related to, because there's a lot of transfers that are paid out basically. And I mean, the reason that they are distributed through time is that they cannot do it all at once. So they do it in a staggered fashion, and they said like, they tried to do it in a reasonable timeframe, but there might be some people who get it later. We can try to play around a bit by shifting the state forward. So, but I think it's not going to make a huge difference for our estimates. In terms of anticipation, I think that's a very valid concern here. So I mean, as Anika said, we tried to do something about that. I wasn't able to show it, but we look at basically the date when this policy was announced, we look at families with children, without children, see whether or not during the next month they increase their spending or not in anticipation of these payments, which we don't find, but there might be more sophisticated anticipation behavior. So, I mean, we can look at our pre-trends, and if there was anticipation going on, they should show up somewhere. So somewhere the people who sort of are closer to the payment should anticipate this more and diverge from the people who are farther away. But of course, we cannot fully exclude this possibility. That's just to be transparent. The aggregate consumption patterns over time, that's also an interesting point. So this is mainly this increase over the second half of the year. This is like a seasonal pattern that you also see in a 2019 nature. So at some point, what we did is we take the, we took the triple difference. So basically like also taking the difference to 2019, this doesn't make a difference for the estimates, they just get a bit noisier. So we decided to drop that, but yeah, so this is mainly driven by some seasonal pattern. We haven't had a look actually, whether this means that people are actually going out more during this time period. They may just be spending more conditional on going out, but that's something interesting we could have a look at. The age of the child, we know. So we can do the check that you talked about. And it's true that if you're still studying, then you can still get a transfer until 25. This is a small percentage of the overall transfers, but we can have a look at it. That's a good point. And yeah, the differences with the number of children, I mean, I totally agree. So I mean, ideally we would like to have like a better experiment where we have different transfer sizes and then we can really tease these things apart. We're sadly a bit constrained by what we have. So I mean, that's, I would argue the best we can do. And I hope I didn't forget any questions, but if I did, I mean, maybe in terms of quantification for inflation, I mean, that's an interesting question. And my priors also that this might have contributed to inflation further down the road because people were just having so much money and then at some point restrictions were all relaxed and people started spending again. So this might have contributed to that. I guess with our identifications, very hard to go so far out in time. And yeah, so I think that's all I wanted to say.