 Good morning and welcome to the second session of this conference. My name is Christoph comes from the ECB We now have a session on the design of fiscal rule and on fiscal consolidations Where we could say there is a common theme on the trade-off between fiscal discipline and macro stabilization Congratulations to the selection committee for choosing two Very nice papers. So both of them deal with classical questions or what I would call quest classical questions dating back to the 1980s So the first paper which will be presented by get ministers and discussed by Facundo Pigalem Basically deals with measuring fiscal discipline, which is of course a theme that has accompanied Europe and European Union since the 1980s and and goes back to questions also asked by the the law committee at that time It is also very topical in light of the ongoing discussions in the ecofin on the reform of the fiscal framework and as you all know, there will be further attempts to Reach consensus on Wednesday The second question likewise is a variant of a classical question asked by Javazzi and Paglano at the end of the 1980s At that time the question was can severe fiscal contractions be Expansionary tales of two small European countries and and you all know it's Denmark and Ireland and Since then there has been a very active literature and so we're glad to have Roberto Pirotti who will present His paper was Luca Salah and also happy to have Gennards Muller from University of tubing and as discussed and Of course, there are even more people in in this room. We will see more people today speaking Who have contributed actively to this literature? I see Sylvia Adania and also I just had a brief chat with Olivier Blanchard who happened to be the discussant of Javazzi and Paglano More than 30 years ago. And so with that, let me give the floor to to help for the first paper. You have 25 minutes Okay, so let me start by saying thank you very much to the organizers for Inviting me and I cannot stand right I need to sit Good If you want to be visible, yes Okay, no, no, that's fine This is a paper about measurement. Okay, it's about measuring fiscal discipline and it's you know partly empirical partly econometric paper with one sort of key new point So let me just a brief introduction. We all know that our fiscal rules Okay, and they are intended to constrain fiscal policy this discretion limit debt financing And at the moment 90 plus countries have implemented some form of fiscal rules And as we see in the data, they are often violated, okay So approximately 50% of the time across countries or across time periods these rules are violated. Okay, and One important feature that I think for why this happens is that fiscal rules depend on endogenous variables Okay, for example, the budget deficit which is affected by factors that are outside of the control of the policy maker So you see here this very nice quote from Blinder who was discussing the ballast budget act in the US Where he said, well, you know as a Congress, you don't really have any control over this budget whatsoever over the deficit really So that brings us to the key question of this paper How do we establish quantify and compare? Responsibility for rule violations, right if the policy maker does not completely control these variables How do we quantify which part is his responsibility and which part is luck? So think about it bad policy bad luck question. Okay, so Let me put this in a picture European context here We see the budget surplus of France and Germany and everything is Smoothly together until approximately 2005 and then they start moving apart Germany stays roughly above the 3% limit and whereas France is nearly continuous below there Okay, the question is does this imply that France is a less Fiscally responsible member of the eurozone or is this just that while the economic conditions in France were worse relative to those of Germany which Creates this difference that you see there. Okay, so that's the question that we want to answer in this paper essentially and Let me give you the main overview the broad idea what we're going to do is propose a Revealed preference approach. Okay, so the main sort of insight is the main idea is that All the ideas it was just said a macro stabilization can conflict with fiscal rules Okay, and the moment where such conflicts arise, right? We can actually measure from the amount of the violation, right? The fiscal discipline or the fiscal efforts made by the policymaker, okay So it is these deviations from Violations of the rule that we are going to exploit to identify The preference of a policymaker for respecting the rules Okay, formally we will define fiscal discipline as the weight placed on fiscal rules relative to macro stabilization when deciding on fiscal policy and Then the big bonus of the papers essentially with that definition We will be able to show that we can compute this notion of fiscal discipline from sufficient statistics Okay, so you do not need to have a fully specified structural macro model in order to learn Estimate these fiscal discipline parameter you can learn it from constructing forecasts which we conventionally do when we do Evaluation of like fiscal rules and you need impulse responses So impulse responses to policy shocks those two sufficient statistic will show are sufficient to identify this parameter Fiscal discipline parameter, which is just the weight placed on the fiscal rules relative to macro stabilization. Okay, so that's the main idea of the Paper and feel free to interrupt at any time if you have any questions. Otherwise becomes a bit static. Okay, so This paper starts with an illustrative example. It's a toy model Okay, it is nothing to do with reality is just to capture this main idea and explain How our approach to measuring this sort of quantity fiscal discipline is different from what was previously done in the literature And obviously some argument on why it might be better. Okay, so the the toy model Has YT the output gap Xt fiscal deficit and there is a fiscal rule the fiscal rule is very simple It's just xt smaller than or equal to x bar t. Okay, x bar is going to be zero for most of the talk. Okay The output gap is determined by the fiscal deficit the deficit and some shock x it think about any External shock affecting the output gap. Okay, then comes sort of like this the determination of the fiscal variable there's going to be two components in there and I would like you to think about it in the context of blinders a Quote right there's a part of the fiscal variable that the policymaker controls and there's a part that he does not control Okay, however, which factors on each part is not important for the purpose of our talk since it's such a simple model I'm going to think about the part that he doesn't control as an automatic stabilizer Okay, this could be due to many different factors But the coefficient here beta is taken as given other things that you could think about maybe there's a risk premium shock Right outside of his control that would affect the deficit for example So you can think of many things for the simplicity of our model We just have an automatic stabilizer there then the other component is the discretionary component This is the component that the policymaker is liable for okay. This is the things that he gets passed through Congress or That he actively decides on okay a policymaker in in the discussion air component in our setting is has two parts There's a purposeful choice Theta, which is the response to the Structural shock and there is maybe he makes some mistakes, you know some mistake abscala and tea some exogenous shock to discretionary policy, okay So that's the model now what we remember what we wanted to achieve we wanted to measure What is the contribution of the policymaker to if the rule gets violated? So suppose that x t gets larger than x bar right? What's the contribution of the? You know the the policymaker to that okay, so Well, what's the first thing that you think of then well? Let me try to estimate the discretionary component right is the most natural thing to do We've just said well, that's what the part that he's responsible for and well. Let's try to get that out So this is known as cyclical adjustments, right? It's a well-known thing. It's done that nearly every policy institution Lots of literature cited there and it's two steps You extract the discretionary components from that model usually that takes the form of estimating beta using a regression and then Looking at the residual which is a discretionary component, okay? And then you compare that to the threshold and you can compare it across time or across countries That's what people kind of do in order to establish the the contribution of the policymaker Okay, so this approach I think this is well known Not very much dealt with this approach is two key problems and they're really like, you know, they kind of bite Okay, the first thing is that is really hard to separate this Discretionary and non-discretionary components from each other empirically. Okay, there's a big end to a genetic issue here In the sense that they are determined by exactly the same shocks, right? They are Simultaneously determined there's potentially a lot of omitted variables because I don't know exactly what he controls and what he doesn't control And there's also a bunch of measurement error problems Okay, so even if you could avoid all of that So imagine that you would be able to extract PT accurately the discretionary component it's still a little bit problematic because It depends on shocks. Okay, so now we can have two policy makers that have an identical reaction function so an identical parameter theta yet their Discretionary adjustment component could be very different because they faced a different shock right different countries get different shocks You know, I have the same notion of fiscal discipline. I want to make the same effort I was just unlucky because I got a different shock as one policymaker. So that's The issues with fiscal adjustment. These are, you know, widely discussed nearly every, you know, like Document that uses cyclical adjustments also describes the limitations of that. So that's that So what we're going to do is not try to fix these problems because these problems are really hard to fix and it's kind of You know, it's kind of nearly impossible because PT is latent and it's endogenous So dealing with latent variables and dodging us and then dodging us is it's hard So we're going to propose a revealed preference approach that is going to sidestep avoid these problems. Okay, and Let me build you up to the to that idea. Okay Consider the following consider policymaker number one Who ignores any fiscal rule? Okay, he doesn't care about the fiscal rule. All he cares about is macro stabilization He's going to minimize the squared output gap subject to the model and then he's going to find this optimal rule He's going to set theta u equal to minus one over alpha. Okay. Now consider the other policymaker This guy is the you know How you say the the the good guy or something or you know this policymaker She will Perfectly follow the rule perfect rule. Okay Okay, you solve the model We get a suppose that it's binding and then you get a constrained optimal policy theta c and then obviously in this scenario The fiscal loss is going to be zero, right? So I just squared the the deviations from the rule and then you have a fiscal loss is equal to zero in this case All right. So now is the key thing what you can note here is that? These are two polar cases either we ignore the rule completely or we perfectly respect the rule and In between there there lies my fiscal preference Okay, there lies my preference for the fiscal rule specifically the unconstrained one pays no attention to the fiscal rule He's going to have a low Stabilization loss low macro loss, but a high fiscal loss and the other guy has exactly the opposite Okay, so he has a high macro cost But essentially very low fiscal cost because he follows the rule So there exists a range between these two parameters theta you and theta c where each theta You know gives you a different preference or fiscal rule stabilization okay, so in different preference for abiding by the rule and the policymaker's choice theta not is Going to reveal this preference. The only thing that we need to do is tease it out We need to tease out from this policymaker's choice. We need to tease out. What is his preference for? fiscal rules, okay, so that's we do here what you can do is you define an auxiliary loss function So lambda is a parameter here that we introduce ourselves. This gives the weight on the fiscal loss Then we can again Solve the policy problem. Okay minimize the loss function now this auxiliary loss function, which is capturing The the preference parameter lambda and we can solve this PT star As a function of the lambda of the preference parameter, then you just have this coefficients for theta lambda, okay? So what can we do now? We can define fiscal discipline This is this fiscal difficult discipline now becomes the difference between the policymaker's choice PT Not what he chose to do and the lambda optimal policy choice Minimized in expectation over lambdas. Okay, so you pull out the preference parameter from the difference between The proposed policy choice in the lambda optimal policy choice, okay? So intuitively lambda not it's just the weight that the policymaker reveals He reveals his preference on average on the fiscal constraints when he chooses the policy PT not, okay? Now the important part is that lambda not is comparable across countries and across time periods Why it's defined a relative well relative to the optimal policy which takes into account the economic environment and takes into account the Shocks, okay, so it's not shock dependent anymore, right? It's just a parameter. Okay? That's the definition of fiscal discipline now We need to sort of find a way to estimate this parameter. So this is the definition It's nice and comparable doesn't suffer from the problems that we mentioned before but we still need to estimate this parameter So, how do we do that? Well in the simple example is not so hard if you put The threshold to zero then you can solve for lambda in a nice analytical function So you have a mapping from lambda to theta not theta not the policy choice maps to a specific preference parameter It's as easy as that However, this approach if I now want to generalize that to a bigger model a realistic model would require that I actually know that model Which is not something that we really are comfortable with Assuming when we want to assign responsibility for rule violations, okay? So instead we're gonna go for a sufficient statistics approach that sort of builds on a previous paper that we wrote You can recognize that the Lambda optimal policy PT lambda is a solution to a linear quadratic problem Okay, in this very simple case this implies immediately that the distance to any proposed policy can be computed from two quantities one forecasts for The output gap and the fiscal deficit under the proposed policy choice So these are the forecast that we normally already compute You know to evaluate whether a policy maker is respecting the rules and the second element are the impulse responses are Impulse responses to the policy shock. So using those two things you can work out You can compute. What is the distance between PT not and PT optimal the lambda the optimal policy and the proposed policy That though that distance can be worked out for any lambda and then you just need to minimize it. So here's the result Well, this is just the expression for lambda not okay It is just Depends on the impulse responses are it depends on the forecast which are captured in YT not and then it's just minimizing this expression here After you replace those guys with Estimates, right? So we still need to estimate the impulse responses We need to plug in an actual forecast and then we minimize the lambda We get it out. Now now I have a problem because I had some beautiful pictures on intuition I have no idea on how I'm going to click on that button at the moment. So How much time do I have? I have 10 minutes left, okay, so let me not give you the intuition of that formula Instead I will you will have to believe me that I just can compute the distance between these two things from impulse responses and forecast And I would like to share with you some of the empirical results that this method proposes us Okay, I hope that the main idea is clear you pull out the preference parameter from the distance between an optimal policy choice That depends on lambda and a proposed policy choice and you can write that as a function of impulse responses and forecast You can do this for any market linear macro DSG model. So I'm not going to go over this, but this result holds for any Generic macro model of the following form. So you have your new Keynesian models in here You have some Hank models in here. You have some other models in here I don't know. There's a lot of them in there. Okay, so this is just a very general model representation Then you pick your macro loss function you pick your fiscal loss function and you define your auxiliary loss you Pull out your fiscal discipline. You get exactly the same formula. I just wrote it in bigger notation So let's ignore that and go to the empirical work. Okay, so I want to illustrate to you this Method right because that's what it is. It's a method for for measuring the fiscal discipline of a particular Country and we're going to do that in the context of the EU in the context of the eurozone and for simplicity We're going to consider a very simple loss function, which is auxiliary loss function. I should say which is going to be square deviations of the output gap for each periods ahead so in expectation and The budget surplus and we're going to take s bar as the 3% threshold from the SGP. Okay, and then we're going to compare lambda nods for different EU countries and time periods Okay, and that's the main idea. So we need two ingredients in order to compute this We need forecasts those we're going to take from the SGP directly as they were delivered to the European Commission So they are conditional on the proposed policy path and the second thing we need impulse responses And then we're going to just browse through the literature pick the impulse responses that we like Here's a slight issue here You see the forecast that were provided to the SGP on the top panel. You see GP growth That's our macro stabilization objective on the bottom panel. You see the budget surplus you see small issue with these forecast There is a slight bias in there Towards that the budget surplus will go to zero It happens You know you get it wrong sometimes so we need to do some Adjustment so what we did we did a bias adjustment here We said we removed the bias for both countries set it on average to zero such that you get something like this, okay? You can do better things here obviously, but this is just to get the main idea across You see clearly in the bottom in the left corner left bottom corner that France was a rather optimistic about their surplus So when we buy a suggest we get something more and there's at least unconditionally over the period unbiased Okay, which I think is a reasonable thing to do second thing We needed impulse responses So these are impulse responses to the fiscal policy shock and here we take the shocks Which you're essentially fiscal austerity shocks they're taken from this paper by Gerhard Lé and Pascatori and you see the usual Shock physical austerity shock goes up GP growth goes down budget surplus stays up So these are the impulse responses that we're going to use in combination with those forecasts in order to tease out the lambda knots There you go So these are left on the on the y-axis is the value of lambda on the x-axis are all the different countries that we considered It's capped at four Let me explain why because if you do not violate right then you put higher and higher value of lambda on the fiscal rule So and if I never violate that parameters, it just goes off to infinity Okay, so therefore we're kept it at four you can just think about they were very very fiscally disciplined What's what's the main finding here while there's a there's a part that is sort of like as expected that Some southern European countries like Portugal Spain, right? They don't show very high fiscal discipline But what is slightly more surprising and which I would say one of the main findings here Is that also countries like France and Belgium which have a considerably better macro growth outlook have a much Lower fiscal discipline as well. So they are definitely in the group of countries with low fiscal discriminant on the right hand side We see Some of the other countries Finland Sweden Netherlands, which all have very high fiscal discipline You see this takes into account uncertainty across forecasts impulse response and everything and sometimes you see that for Greece For example in the middle the uncertainty is very very high. So we just get very very large confidence bands on what their fiscal preference was Okay, so that's one so let's look over time now So on the left-hand side, we see the start of the euro, right and everybody was perfectly fiscally responsible So it all went great Then already sort of immediately after the the the euro was initiated You see a big drop already in fiscal discipline and not as well that the variance Increases massively so that means that there's a number of countries that are really You know pulling this fiscal discipline parameter the lambda parameter down They are becoming much less fiscally responsible even immediately after the euro was started and the same It continues a little bit and then you see a rise again in the fiscal discipline parameter, okay So we'll come back to that first observation in a second to sort of Rationalize like which of the countries were pulling that And here we are already So here you see France versus Germany So the first thing to note is that both France and Germany were kind of responsible for this large EUI drop in fiscal responsibility immediately after sort of the euro started Okay, so both of them are on the lower point And we can also see that answering the question that was first posed in the papers that indeed France is actually Fiscally less responsible. So what's going on the macro growth outlook of France is actually pretty similar, sorry, it's similar when compared to Germany only the Fiscal deficit forecast is much much worse. Okay, so they well what what is revealed is that they have a much lower preference for fiscal discipline and you see that that continues all throughout the period that France is significantly below Germany and also significantly below the majority of the euro zone in terms of fiscal Discipline Okay, so let me summarize Defining measuring fiscal discipline is hard because of these two big problems comparability endogeneity We propose this new approach based on revealed preferences. It's easy to implement using sufficient statistics It's well identified only requiring forecast, which you already have an impulse response functions. That's it Thank you very much. I give you a four on your time keeping discipline So now it turns to Facundo you have 15 minutes We're working in similar things. Oh, sorry So it's nice. So it's different for me. He's an econometrician. I like to do theory. So It's a nice complementary that was So what is the paper about and I can do a story at the beginning when the environment they sent me another version of the paper Uh, which was quite different This is not a complaint because when I read the other paper I thought I would write a completely different paper that in that being this So when they sent me the new version, I was very happy So it's not a complaint. Uh, so what what is the paper about? Many times the fiscal rule are not respected and we know that And but we don't know as he said we don't know if if it is mistakes or or is because the governments are doing on purpose, right? So it's complicated and I remember when I was young I used to do econometrics And and I remember dealing with the data you say, what is GDP? suppose the european union is The the deficit limits 3% you said what what is the GDP? You know, what time are you measuring? What is government spending? What kind of government spending? And and how is it changing over time? I remember doing for it because they were wrong at the beginning when you get the right data five years later They were correct And so all these issues can imply violations that are unintended But but then the governments can use these like yeah these mistakes to purposely Break the rules and apparently france is doing something like that with some bias estimates and So what they do is they use a rather The the revealed preference approach, right? And the idea is to measure how intentional Are these violations And and then you try to find the way you assume there is some some weight between like doing your own discretion What is good for you and what is the? The the the the cost of violating the rule So what I'm going to talk about from the theory point of view is at the end this Discipline is measured as a different respect to an optimal policy. What would be the optimal response? to to the particular shocks so then What matters are the details of those models and what are the payoff functions? Why how do you interpret what you find and another thing? What could go wrong right because you're You may have it wrong What is the optimal because you're missing you have a different payoff function You have different model or something like that and so something can go wrong We all write papers that something goes wrong. That is fine. One just need to know what what can be wrong so and let me come back to the to the To the simple model the toy model and I just put in the Replicating here. Do you see in there is the minimum minimize in the output loss? and The break breaking the rule only when you break it. That's why there is a plus sign in there subject to some dynamics of the deficit and the and the output So you see here in the toy model He puts it like like this But in the in the main model there is an expected value ex ante And then you have to think this is important also because you you think when you are Optimizing you know the shock or you have to decide it in advance And I don't know exactly what it is You know because you you may think look I I'm deciding the budget in 2023 But for 2024 so whatever So I'm committed in some way to whatever is going to happen in 2024 But it could be also that you you see the shock In 2024 and you re-optimize the policy So it's not clear to me if there should be an expectation in there or not I'm going to do it without the expectation But I think in the paper they do both and they have with expectation. So there is a small detail So and here the important thing is that the the station you see the x there is a discretionary component And there is this stabilizer component And when I read the paper the first thing that I thought at the beginning is for me was all all about beta It was nothing about p I you know in argentina as I said before and if you think about the fiscal crisis in argentina is You know the the former government created like three million new right that is Two million new, you know social assistance plans that give you like five million people They have no discretion to spend so pt is fixed But it's all about the choice of the beta that they make and and that obviously they have that response to the shock Uh, that's what I thought Uh, but it's okay. So I would think in the future about that It's not about uh, not as much about the discretion But also how you come in ex ante to do something coming back to the france example. I remember I was there Three or four months ago. There was all this protest with the With the uh pension reform And you may think that that is targeting the beta And not p and and so the idea is to understand what happens when you Focus on p and and and maybe you don't you don't focus on beta Okay, so let me do them also. I thought what would happen You know, what how beta matters and I said well first the probability of breaking the rule I computed is You have a low shock You break the rule and depends on the fiscal multiplier doesn't depend on beta Okay, then I did the output and then the output you can the variation of the output Um, uh, you can stabilize it if the rule fully stabilize it if the rule you have a sufficient space If you have a big shock you break you you're hitting the the constraint in the in the fiscal rule You cannot stabilize it again doesn't depend on beta and what happened with the deficit the same Nothing depends on beta in this model Um, and you you wonder why You know when you I thought ex ante I thought ex ante. This is super important Long term long term structural reforms, but in this model apparently doesn't matter Is it the toy model? I don't think so and what happens is like P is so restricted In in when you compute the optimal the optimal discretion. So you don't have limits on p So in the margin the only thing that matters Is is your discretion of response because you can do whatever you want Just give me one beta and it's just I can adjust the p to get exactly the fiscal outcome that I want Okay, and uh, and uh, and this I think is going to be general It's going to extend also to the general model even besides the the the toy model Okay So and then I thought okay. What what I would I do he's an econometric He knows much better than me how to do these things, but I thought how we'll do this and Suppose you do a structural estimation you can do it Uh Then you have to do an assumption about the models how the things are moving And and then you can potentially identify And here you see you I have four parameters And two four estimated parameters are four fundamental parameters. So I can easily recover lambda from these Dynamics if I do a structural VR and this is also revealed preference in some way, right? So I have to assume something about the model And and that is fine Of course in this simple model, I it's over identifying the general model You're going to have a lot more parameters. It's going to be a lot more complicated So I don't want to go into those details and I I I have no idea about the the potential Statistical properties of of those estimates But potentially we can do it and it will be interesting to see If an approach like this give you something consistent to the approach That they are using in the paper, you know, hopefully you're doing different ways and they all give you Similar signals about lambda and that will be reassuring About what you're estimating for the physical discipline Okay, um Okay, and so then what the autos do and then I thought okay, how better matters you see I have a slightly different equations What I put here is They they assume that x for the solutions are x bar at the limit is zero and then a a big chunk disappears in there But my point here this goes to the ex ante The ex ante and exposed When you are choosing because this response may be state dependent When you are close to the physical limit to the deficit limit, your response is going to be smaller Definitely, right? So because you know, you know, even as long as lambda is not zero So if you really care about you care something about satisfying this rule Your response is going to change and that's going to create an state dependent parameter That can create some potential complications Especially when you do it ex ante when you do it expose and and as I said, it's not obvious to me If what is the best way? Are the governments choosing fully ex ante or fully exposed? Probably it's uh, uh, it's a combination of the two and so just to make it shorter and uh, oh So what do I found? So lambda is decreasing in beta, you know, you can do you can do the in this storage model you can do some some Some math in here and even though Beta does not seem to affect the system at all It does potentially affect the estimation because how you separate the discretionary component from the other I don't know if that is good or bad I I don't know, uh, but it's going to be very relevant Uh, what is it on the line in the underlying beta in the model? What is the structural stand of the economy? I I don't know. Um Yes, so even if you can change beta in this model, it doesn't change the probability that you break the rule but I don't know So that's it. Let me let me go back Let me say some quick comments at the end as I said, I like much better the new version It's exactly the paper that I would like to write But I I think thinking about When you think of a reveal preference, especially when you're taking a distance respects to an optimal You have to think exactly what what are the necessary components that you should put to make it more believable Uh, first I tried to do a different constraint because in the model, you know, it's like if you think about the Deficit to GDP ratio and you put in locks You should do x minus y minus Some limit It doesn't change the parameters are a bit different, but it's the all the results that they have is exactly Uh, but then you start to wonder what are the payoff? Of course, you you need to keep quadratic loss functions um Because of statistical properties you take the first order conditions are linear Everything is linear and everything is nice Uh, but I was wondering what are the garments trying to do? You know first is maybe the garments are trying to smooth the garment spending I don't know if they are actually concerned about Smoothing the output because that is complicated But maybe you don't want to create the austerity measures You want to keep your you know the the wage bill more or less constant You don't want to affect the spending and so on maybe this is a potential loss function I will think more about limited discretion In as I said in the model when you solve for the optimal lambda is you can do whatever you want you can go You can pick p at any level that you want and you may think there are some constraints in the economy on how much you Have far you can go and that potentially it may change the results So coming back to the france or or germany, I don't know. Maybe in france is harder To adjust the the the fiscal deficit. Maybe I don't know So if there are french and german people here, please let me know but uh But you may think that that is that could be relevant when you are where you are doing the output loss uh, and the other thing is I don't know how to think about the default risk It's supposed to if you see I think italy. I was looking at italy. It doesn't look that bad right and uh And I don't know exactly you may thought that maybe italy is worse But there is some kind of discipline imposed by the market because you know that you increase your government spending You know the risk premium may increase and that is it may look like a lambda And in the in the model Okay, so I don't know so maybe I will try to have some lost function about that How much do we care about the risk premium? You know, you may think I don't know he can tell me better. You can think like maybe germany are more It's more concerned about increasing the risk premium the default risk about the sustainability of that than the frances And that may look like the germany cares more about satisfying the rules than france Uh, so I don't know so these are as I said, I have more this a new version. I think it's like a new draft so I tried to to to give you things the directions to what witcher would like to To include in the model, especially if you are computing a deviation respect to an optimal policy You need to think seriously about How are you constructing that optimal policy and what are how did the details Matter, right? I didn't find all the components and that's it. I thought I did great with the time Do I do I get a four to you get a four? That was just asking whether that's good or bad It's very good the floor out of 10 So thanks a lot for the great presentation and discussion. I mean we have time for Like three questions from the floor. So let me start with Giovanni and then uh, vitore and Christian Thank you very much for both the both the presentation and the discussion both very interesting One question is on on what we really mean about discipline. I was also um, surprised about the results and To me was driven by the fact that discipline here is about the deficit the budget balance discipline It's not about the debt because if we look and the sample lots matters because it's 98 forward Italy was already coming in with a very high Debt and then the response is Is coming through as well the other one. So this is more Indefinitional terms. The other one is more more in in in your way of defining the The the the distance between PT and preserved, which I liked a lot. I don't know your paper. You are paper Regis, but I wonder because if I think about the the stabilization function beta that we were talking about I could see two different types know the automatic stabilizer, which are immediate And the systematic component that like branch out and priority Totals in the past it comes with luck Is this difference in time going to matter in the way? This this this policy is backed up In the sense the difference in timing, which is very important in when it's something that also the discussion was raised. Thank you uh, thanks a lot for the uh paper and for the discussion the two questions I have Have been covered by fecundo, but I want in a sense to give that my own twist One thing that I was struggling during your presentation guard is how does one think about constraints on policy? That do not come from the rules One of the things that I would take for granted is that public spending Cannot jump up and down easily In particular, it cannot come down very quickly And that may be something that you will want the other thing is when I was when you were presenting I was struggling with this idea of how to take into account the endogenous the expectations and market discipline Right, but I was thinking that in your setup that should be taking into account by the model That you're implicitly using but then I got Dervis about this interpretation. Please help me And I wanted to follow up on the the connection between your empirics and then the the infinite horizon theory in the following sense So the definition of your lambda Part of like ingredient to get at this is the optimal choice given a lambda In the infinite horizon case what you'll need to know for this one as usual is for every like say your fiscal instrument that you're adjusting some Say government government purchases you will need to know the effects of every possible path of government purchases You would be adjusting now the connection with the empirics where the question is because we don't observe this We just observe certain kinds of historically observed fiscal adjustments. How do I know that the differences I get at the end? I actually Differences in the the true lambda Which is something I can also define if I want to know all of the causal effects of fiscal interventions versus Across different countries. I've just seen different paths of fiscal adjustment At least then I thought I couldn't tell those two apart Thank you. Okay. No, I will start with Facundo. I'll pause like through all of them so, um The role of um beta in this model is just purely illustrator. It was to illustrate that Cyclical adjustments are difficult, right, which motivates our parole Facundo is completely right. It plays no other role for what we do whatsoever. Okay So in the general model, you will see that the discretionary policy depends on all endogenous variables all everything in there We might actually put that back in the simple model to actually emphasize that that is indeed not something that is driving any of these results here. Okay so, um You can learn lambda From estimating the model. Okay, it's just that that requires specifying the non policy block And it requires taking a stance on what is the policy rule So we just simply have a preferred approach of learning it from sufficient statistics Okay, that is a preference because I think it's You know, it works for a broader class of models, right? You require forecast impulse responses That is a preference if you will as opposed to estimating a full-fledged structural model um Our definition of fiscal discipline is unconditional Okay, so it is an expected value of this difference between these two things in facundo's thing. It takes a conditional definition of fiscal discipline in which the lambda parameter ends up depending on the structural shock. Okay You can do that, but What happens then is that Is essentially not comparable again any more across countries or time periods, right? Because the shocks are going to vary across time in countries making the lambdas difficult to Compare like one of our objectives was to have a measure that is comparable across countries and time periods Hence the unconditional definition of fiscal discipline um Several questions pointed to what do you do when there are additional constraints that um that the policy maker needs to satisfy Okay, so um in this aER paper we deal with that issue and to give you just a broad flavor of how you deal with that You replace the loss function that you have By the lagrangian that takes into account the constraints that you want to impose So any additional constraint that you want to impose on the loss function You impose it by imposing it on the lagrangian and then you compute the optimal policy as as as as normal Okay, that has does not have a neat analytical solution anymore, but and this is crucial It still depends only on sufficient statistics. Okay, so it still depends on only the impulse responses And the forecast if you stay within this general linear model class. Okay, so that uh takes those uh question So then there's a question. Do you need to use a quadratic loss function? No, you don't Is it convenient? Yes, of course If you do not use a quadratic loss function, you do not you don't need to use you get a still sufficient statistics But they're not in terms of the point forecast anymore There you're going to need the density forecast And you're going to need other functions of the forecast distribution in order to evaluate your statistic And that can be considerably harder in practice. Okay, that is not a given that you can just do that easily. Okay um That that that that depth. Yes, we made a choice on in the empirics which variable to impose So we impose the budget deficit, but it's a great idea to think about like what if we would impose depth whether the Results would change absolutely fair fair fair that that we should definitely do. Um How does That was the constraint question And indeed so then if I have everybody's to have all the points I then I will talk to talk about christian's point like so It is indeed true here that you need for a full fledged evaluation of fiscal discipline You would need to know the impulse responses to all fiscal policy interventions And it is indeed a question like um, uh, you will have to say that My interpret so suppose you do not get it. Suppose you just do what we do and you have just one Impulse response You will have to make your definition of what is fiscal responsibility Limited to the interpretation of that particular instrument. Okay, so it is just a limitation of your Explanation, so you just have a different interpretation of what is that coefficient? But yes, absolutely like ideally I would measure all possible policies that it could contain And then based the computation of lambda based on that on the revealed preference, but in practice We will be limited to a uh finite subset of possible interventions that we can Measure I think my my personal view from the fiscal literature in the last sort of empirical part in the last 20 years that people became much better At estimating causal effects of impulse responses. So I think there there's a lot possible there already also because You know like yeah, there's just a lot of literature that we can draw from to improve that part of the of the paper Please forgive me if I forgot any question We have the brakes also to continue The discussion. So thank you very much to both of you If you can pass me the So that was that we turn to the second paper and I'm very glad to have Robert of the Rotary Thank you. So this is a paper with Luca Salah also at Bocconi University So here in this paper, we revisit the literature on fiscal consolidations. It's a purely empirical paper and there is a sizeable literature on this and Uh The more there's a recurrent finding in this literature that Those consolidations that are done by cutting spending Government spending tend to be expansion. It tend to be to have positive effects on gdp I think probably the earliest contribution is the paper that was mentioned before by javazzi and pagano on the island and Denmark in the mid 80s We mentioned a paper also by one of the quarters of the Current paper the paper I'm presenting today not because we want to be self referential But because we don't want to hide the fact that at least one author of this paper today's paper had different views in the past And they might the recent contribution probably the most influential one is several papers by lezina favo and javazzi including a book That came out a few years ago. That was fairly influential both in academic and policy circles So we start with a panel of 14 countries and uh, it's um based initially on the work by The vries and others at the mf and but then was it was updated by lezina favo and javazzi 2019 ways to use exactly the same data set And as we'll see we we'll use exactly the same specification with one minus change Um There are two key variables that you have to keep in mind These are the only two variables that you need to keep in mind One is a ut is the surprise change in the current surface. This is a narrative the data set. So this is data change Surprises to the surplus constructed from the budget documents So the first variable is surprise changes to the current surplus and the second and that's ut At is surprise announcement about future surpluses So these are the only two variables that you need to keep in mind With these variables, you can define two types of consolidation You can define a dummy variable for two types of consolidations The first one is it which is tax based consolidations those consolidations that are implemented mostly by cutting taxes And i.e is expenditure based consolidations and these are constructing using the current surplus shocks Plus the shock to the next five years the announcement About the next five years you put them together and you construct an index of whether the current the consolidation Today in the next five years is relies mostly on tax increases or on expenditure cats So here we start from exactly from the specification of a la zina fabo and jab atzy and in blue So it has two components. It has moving at the movie average representation And the fiscal plan the movie average representation to think of these years The growth you can be any endogenous variable as we'll see Is a function of lag zero to four of the two variables i mentioned before And The surprise to the current surplus ut and the announcements the surprise announcements about future surplus is at So these are like zero to four of these variables and you allow for a different coefficient depending on whether These are expenditure consolidations or tax consolidations in blue. You have the expenditure consolidations and in red You have the tax consolidations But they are exactly the same And Except that we allow for different coefficients, of course And then you have a fiscal plan that links the announcement about the future to the current shock And so it's posits a systematic relationship between the two variables and again It allows for a different coefficient depending on whether it's a tax based or expenditure based consolidation and Then you can think of an experiment in which you shock ut in an expenditure based consolidation or in a tax based consolidation And it has effect directly through the movie average representation because it depends on like zero to four of the ut itself Indirectly through the fiscal plan the second equation that enters also at the movie average representation So when you combine the two and you estimate the impulse response to a surprise shock to to the surplus You get these impulse responses On impact and up to four years And these are taken again from alizina fiber and jabrati essentially In expenditure based consolidations, which are the blue impulse response You don't get any movement in gdp With tax based consolidations. You get a very large negative effect on gdp up to two point negative two point percentage points um in At four years so Can you go back to so notice that in in this specification that The announcements are linear functions of the surprise shocks. So one thing you can do is To estimate the reduced form in which you replace the announcement with the With the surprise shocks, so it becomes a Irrelation just because between the Endogenous variable and the surprise shocks. You don't need to add data on announcements, which are difficult to get from budget Documents and also uncertain and subject to a lot of Judgment calls You just need data on surprise shocks this year And so this is the reduced form. It's It's a much simpler a much simpler expression And you get one more one benefit of that is that you also get point estimates and the standard errors of the coefficients of the impulse responses without any need to To to do a bootstrapping anything you just get standard error straight from the ols estimates This is just a small detail, but it will be important later and Before you you get to we get to these estimates notice that there are two biases in this approach One is a sensor in bias because the data set constructed by the mf and by adesina farman javasi Is a data set the sensor of the data the In the negative surplus changes to zero all Country years, which do not exhibit the consolidations are sensor to zero set to zero And there is a common intercept bias because notice that the the Intercept c is common to both tax-based and expensive base consolidations And that introduces a bias as we see in a second so Here is the Graphical representation of the sensor in bias. So the two squares to the left of the y-axis The two Changes in the surplus which are negative, but in the data set the sensor to zero And they become the crosses and in the the original Estimate estimate the interpolate the or less interpolate through all the Squares to the right of the y-axis and the crosses and that obviously introduces a bias And the and the after the estimated the or less interpolant is the broken line, which is intend to have a A bigger slope than the than the two interpolant This is fairly easy to eliminate just eliminate the sensor data. It's not you can't eliminate all of them in a dynamic setting because you need to estimate the The the lags, but you can estimate you can eliminate some of them reduce the bias And the common intercept bias is here. So the solid lines are the actual or less interpolants in the tax-based and the Expanded to base consolidations the red and the black and the blue without once the broken lines are what you get from Imposing a common intercept And you can see that in the case of the red line you can even get in the opposite slope than the true one In general it tends to increase algebraically the slope of the Of the Interpolant with the higher intercept and decrease algebraically the slope of the interpolant with the lower intercept And you can easily can easily get rid of these bias by Allowing for a different intercept of both of the two types of consolidations So here is what we get when we We estimate this reduced form which is implied by their model but with this by allowing for a Different intercept for the two cases and Eliminating or reducing the censoring bias and we get the same qualitative results a large decline in gdp In response to a surplus shock a tax-based consolidation And virtually zero response of gdp in response to an expenditure base consolidation So so far we get the same result which is good although with a reduced form which is Easier to interpret the next thing we do is to actually estimate the actual responses of actual government spending and revenues to the surplus shocks the negative surplus shocks And here is what we get initially. So this is The impulse response from the specification of alizina fava and javazzi with the same with one intercept With the whole the whole data set which includes a lot of zeros which is subject to both types of bias And their definition of government spending which is government spending over gdp Which is endogenous to gdp because if gdp falls government spending over gdp increases a lot So you see that here it's It's There is not much difference between the two if anything It's for government spending falls more in tax-based consolidations But this is not statistically significant the next slide We estimate the response of actual government spending again. That's important as opposed to announced government spending And and here we allow for different intercepts. So we eliminate the common intercept bias We reduce the common the censoring bias by Using a smaller data set eliminating the zeros that we can eliminate and we use this dependent variable the log of real government spending which we transform as a shelf gdp by multiplying by the Average shelf government spending to gdp. So this is not influenced by movements in gdp And you see here that so this is government purchases. So it's the constant government consumption plus investment Which is not doesn't have any built-in response to gdp through automatic stabilizers you see that here What is labeled as expending to? base consolidations, which is the blue line And exhibits no change in expenditure While what is labeled in the data set as tax-based consolidation the red impulse response exhibits A large decline very large declining government purchases government purchases is typically 20% of gdp And it's about half of the total government primary spending The original shock is 1% of gdp And you see that at three years after three years It's all eaten up by the declining government spending in tax-based consolidation. What is labeled as tax-based consolidation? The next slide Is sick adjusted primary spending. So we take all primary spending including transfers We and transfers have a large endogenous component large automatic response to gdp. So we Get rid of that by using the elasticity on transfers to gdp computed by the ecd And again, we get the same result in what is labeled as tax-based consolidations government spending declines a lot And what is labeled as expending to base consolidation of the blue line? Government spending doesn't move at all And the difference is statistically significant the 10% level as here like here or even at the 5% level Remus Exhibit if anything the opposite pattern Although then the difference is not as statistically significant In tax-based consolidations. They what what is labeled as tax-based consolidation? They decline a bit in Actually, they essentially don't move what is labeled as expending to base consolidation. They if anything they increase a bit So these results turn out to be robust in several dimensions We eliminated one country at a time because with this data you always have to be careful that there isn't a Country that has done a large consolidation that influences the results We have estimated the the influence responsive by a completely different method, which is a state dependent local projections And we get this essentially the same results so If you put these things together these results together with the result on gdp essentially This says that consolidations that are based on actual expending cuts, which are labeled as tax-based consolidations in In the literature associated with large declines in gdp while consolidations that are based mostly on actual tax increases that they're labeled Expended to base consolidations in the literature Associated with virtually no change in gdp. So this is the opposite exactly the opposite message To that of the literature on a fiscal expansion and fiscal consolidations So one interpretation of these results is that this expansion fiscal consolidations are either rare or hard to detect or non-existent Another interpretation, which is not mutually exclusive is that the Problems with the narrative measures, which are nobody's fault because narrative measures are difficult to extract from the data You need to look at each individual budget and supplementary budgets And make a lot of judgment calls about what they imply for the budget For the actual going spending revenues But one important interpretation is that I think it's consistent with the experience of many countries, particularly in europe is that it's very popular to announce That you will do an expenditure base consolidation that you will get rid of wasteful government spending put And then when it comes to actually and you announced that in the next three years not now But I promise you that in the next three years I will do that And then when it actually comes to the times that for the times come from For for for the doing the actual consolidation, you realize that it's very difficult to get to cut Government employment is very difficult to cut patients And you resort to tax increases instead One hint towards this is that the tax-based consolidations that at least in europe tend to be according to Our experience of many of many countries in europe tend to be the norm Only one third of the total in these database, which is again based on announcements in the budget Another hint is that if you take perhaps the biggest fiscal consolidation expenditure base consolidation of all which is finland in the 90s Well, the government announced a cut in government spending over three or four years between 92 and 96 97 Of almost 12 percent of GDP, which is a huge number And then if you look at What it actually did and if you look at the supplementary budget that it was forced to do because Every time that it announced a cut after a few months It would have to enact a supplementary budget to essentially renege on these cuts It turns out it turns out that It's mostly it was mostly based on taxes, but this was not captured in the narrative narrative data set So these are two possible interpretations of this data. We need to do more work on that but The impulse response is the raw data that the raw impulse responses seem to be as I said pretty Robust, thank you Thank you very much. Roberto then we turn to gernot's discussion Thanks Okay, thanks very much for giving me the opportunity to discuss robato's paper It's a particular pleasure because robato was my phd. It's wise as i'm 20 years back at the e y And I think uh, this is an important paper It takes issue with that popular or widespread notion that fiscal consolidations which are based on expenditure cuts are Expansionary or at least not particularly recessionary It's also an interesting paper because as it dissects this earlier work by alessina Fabio and ciavace ciavace or afg let's say You know, it's quite a bit of a crime story if you want because there comes these interesting results and we can Try to understand how these results come about I mean on the methodological side you have these three aspects Which are very interesting. So I mean there's this aspect about fiscal plans which in afg Is uh modeling accounted for in this two-step procedure But then as uh lucan roberto show in the end you don't need this if these plans Are essentially only a function of today's fiscal surprises. There's no need to model this explicitly So in the end we are just looking at fiscal surprises and that makes the discussion somewhat more transparent And then there are these two other Methodological aspects which they highlight the the bias due to the censoring of the sample and I come back to that and then The bias due to the common intercept assumption Now given that these methodological aspects are fixed Then the paper comes up with these very interesting and uh provocative results That a there is actually no consolidation whatsoever in what is labeled expenditure based consolidations and second When we look at text based consolidations, they feature only spending cuts And they are still very contractionary But they are mislabeled So so here's a table from the paper. Now you have these nicer figures But I reproduce you the tables. Maybe that's good catering to different tastes So this summarizes the gdp effects from the left to the right left is afg specification right is the Specification which is corrected for the two biases And the important aspect is in the In both panels, you see that in response to eb that would be uh column one and column four and column seven You see the response of real gdp and there is none consistent with that notion that expenditure based consolidation are not particularly harmful Yeah, but then and the second column or the eighth column which corrects for the biases you see then that The so-called text based consolidation are very harmful if uh, you're correct for these two biases They are not as harmful as in their baseline afg, but but still the effect the adverse effect on gdp is quite sizable Okay, so that's More or less confirming The earlier work so but the real puzzle is this then because here you see again What roberto showed you namely that When you look at what's happening to your fiscal instruments during these both During these kind of episodes In the left panel, you have so-called expenditure based consolidation where really nothing is happening to spending Nothing is happening to revenues. You know to be a bit rough here While if you look at text based consolidation, you see nothing is happening to revenues either These are cyclically adjusted revenues while spending takes a takes a bit Cut. Yeah, so this is the mislever thing and then I think the the the question which is Which I want to ask which is natural to ask is Why do these cyclically adjusted revenues not rise during text based consolidations? Yeah, so that's one of two questions I want to address in my discussion and roberto mentions this possibility that In actual consolidation episode it might be politically difficult to to carry out Announced spending cuts and you may resort to raising taxes instead But that doesn't really explain why revenues don't go up in text based consolidation If anything if politically it's easier to raise taxes Then you would expect taxes to go up during these text based consolidations, but they don't So assuming that the cyclical adjustment is correct. There are two possibilities I believe so one is a failure to implement fiscal plans And the other possibility is that the narrative record or the classification by afg is wrong. Yeah And I was struggling with this a little bit and I was looking at a literature I came across this interesting paper by doubler norris and lima They use more reason work at the imf a fiscal reform database Which is Starting from this earlier work by the freeze it out and afg But it offers some more details on the tech side It operates at quarterly frequency and importantly it also distinguishes between announcement and implementation dates And just to give you a chest of the progress which is made there You know, let's compare for a second what you get in afg in terms of this narrative ex ante classifications and I just took Two examples to get a sense of How that looks like so the first one is from germany 91 that is classified by by afg as text based And the text, you know, it's hard to digest It's a narrative record, but it mentions a vat increase in germany So that was easy for me to check so I checked indeed vat went up by one percentage point in 93 s announced in 91 So that conforms with that notion that it's a text based consolidation But then there's this other episode in afg, which is the uk in 2010 which is classified as expenditure based This is the incoming cameron government at the time they have big plans on consolidation And then you get a long text. It's really hard to swallow and why it's classified as expenditure based Is it's hard to see so I found it refreshing or interesting that in this more recent work by dabbler norris and lima They come up with for that specific episode A much more detailed and richer picture And what is striking is that in 2010 in the uk you not only had this spending cuts Planned by cameron, but you also saw a sizable vat increase by two and a half percent from 17.5 to 20 percentage point And what's also interesting to note in these columns you have the announcement date and the implementation date So this was announced in june 2010 and it was implemented in january 2011. So it did happen Yeah, so we have something which is classified by afg as a expenditure based consolidation Which in fact involved this massive increase in texas. Yeah, and we see it's implemented incidentally at dabbler norris lima They also have information about How how long it takes usually to to implement these things and these texts to tech stuff is implemented within a year or so so So things do happen Now dabbler norris lima. They also estimate the response of various levels to these text changes and they instrument those with the Narratively identified text shocks also measuring these by the two year intended revenue In fact, so it's close to what afg wants to do, but they focus on tech shock not on Tech space consolidations if you want, that's the difference they avoid this classification But go directly for the effect of texas and that's what you see So that's kind of reassuring in the left panel You have the response of texas and they do to go up output faults So multiplier being about a one so that's that it's not crazy It's reasonable results, but the important thing is texas to go up in response to this So in light of this my conclusion my take on this is really consistent with what roberti was saying this this narrative x Ante classification doesn't work Yeah, so and then I'm asking myself. Okay, if that doesn't work, why do we do we do it? I mean, can't we just go? Looking at conventional estimates of how different physical instruments work And what came to my mind then was okay, maybe physical consolidations are special and that's why we don't want to rely on Your traditional off-the-shelf fiscal multiplier estimate And one important aspect of consolidations is of course that you look at cuts in spending or tex hikes So but then if that's the concern you you might be fine by just controlling for the sign of the physical shock And as it happens, I have a paper on this with born das khani on paper Uh, and so in in that paper, we have a small model of a Small open economy with where the key feature is downward nominal rates of GDT And a direct implication as you easily see is that The effects of government spending depends Fundamentally on their sign if you raise government spending, there's almost no effect on output You appreciate the real exchange rate if you cut government spending you reduce output And there's no effect on the real exchange rate and it maps easily nicely in the in the in the graph which roberto presented on the Censoring bias so our model that would be the blue line here with the downward nominal rate GDT So you is a fiscal surplus surprise in our case a government spending cut that has a big adverse effect On gdp. That would be the delta set the dependent variable But if you raise government spending if you go to the left of the zero then it only has a very mild Expansion effect on output Yeah, and that's censored in afg so I also use roberto's notation. That would be the the red crosses Yeah, which you know when the actual observations would be far to the left You you have them on the on the vertical axis and then you overestimate the the expect the effect of contractions Not so if you do our approach in born it out We can just estimate a non-linear model which accounts for the different slopes to the right and the left of zero And that's what we do in that paper So we estimate the simple local projections where we throw in Where we allow fiscal innovations to have different effects depending on their sign in the second step We identify them in a first step using sign restrictions Based on Christophe's earlier work, but it works very similar for blotch and perot if you want It also is also robust if we do this in one step and stuff So what we get out of this is this picture In the left panel you have the effect of spending cuts, but now conditioning on the sign So the blue line would be a spending cut and you see it's very much Recessionary that's what you see in the right panel when output falls by about 2% So that's a multiplier of between 1 and 2 If spending instead is raised it has no effect on output whatsoever the multiplier is zero Yeah, so uh, then this confirms well with roberto's finding according to which these spending cuts are very much recessionary We don't have tax tax cuts in the paper explicitly, but we pulled it out for this discussion and also there interestingly We have quite some important non-linearities if you Cut texas then output increases by some 4 so that's at the high end of our multipliers for texas But it also confirms with early literature because if we look at tax hikes They're in fact not as recessionary here the tax multiplier is just a minus one. Okay, so let me conclude here I think this is an important paper. It identifies. Okay, or whether was of course much more polite in putting this so Um, these are his colleagues. I guess after all so he identifies these major flaws in the influential work of afg It illustrates this methodological Challenges when focusing on these colonization episodes But I think what's clear here is that we should be very careful using this tax-based expenditure based distinction and practice And the policy message is likewise very important because it shows clearly that these spending-based consolidations are no panacea. Thank you Thank you very much. Gernot again. We have time actually a bit more for a number of questions. So please Just a question Whether did did you get any reactions from say color faraday? Not yet in the sense that he hasn't read the paper carefully yet, but we are still talking to each other future Uh, more than a question is A suggestion since there seems to be an issue with the with the data set and what actually the narrative approach Really shows I was wondering whether one could use data which have been Put together by the european commission for about 20 years now, but I should double check Which use a what we call a semi narrative approach. That is we look That on the expenditure side one looks at primary expenditure is excluding a measure of cyclical cyclical expenditure essentially unemployment benefit Because the idea is that apart from that this is expenditure that is under the control of the government And whereas we have a narrative measure verified with member states Of discretionary revenue measures. So that may perhaps Help shed some light on this issue of what the narrative approach really shows Klaus the other question Just a small observation on on the last slide of the discussant. Um, I mean You find that spending cuts Are detrimental to growth spending increases not what that does this imply for intertemporal for longer term policies So if you in the past have increased spending too much that's in the long run detrimental if you have to take it back No, but on the taxes. It seems a bit other other way around Yeah, so, um, I have one question the we It's not the the european commission data. Uh, we try to use it. Um, it's available only for the last 10 years publicly and It's so we couldn't we couldn't actually use that and It's mostly for taxes as I understand And so for these two reasons we we couldn't we couldn't use that So we are we are aware of the existence of this data. We tried to use them, but we couldn't So the other question is for you, I think Yeah, so, I mean Of course, this is yeah, so let me put it this way. I mean in the paper we do more things and Here I gave you the very simple picture You also have then we do further conditioning And then it depends a little bit on where the economy is in terms of the business cycle And so this the effects I show you here are basically when the economy is In a in a boom situation then these spending cuts are harmful They are also harmful in a recession But I mean Raising government spending in a recession can be beneficial and then we don't have that kind of symmetry which I showed here So that that makes it maybe more plausible as a result Any further questions So maybe I would have one question if This data set it's not perfect as you've demonstrated, but when it was constructed then Still it was super like a labor intensive task But now we have all those advancements on the large language models front And possibly this could be used to supplement to improve classifications to More precisely determine what were the expectations about those fiscal plans and measures Did you did you look into Into this whether this technology could be applied Like gpt and large language models the textual So given so what I don't know if I can if I the answer will be pertinent, but We spent about one month literally one month on Reconstructing the finish experience for four or five years And So I'm not familiar with the artificial intelligence. I mean the The background of artificial intelligence so But I would say that given our experience with the five years five years of finished data One month of full-time work and Chart gpt would have to be very very very Insightful and confident of what it's doing to To to to give us something that is usable Uh, so To give you an idea So to reconstruct the finished data you have to go to the budget the yearly budget And then in some years there are six supplementary budgets So either chart gpt is very clever very clever, or I don't know how it can do that anyone else I mean think about I would have one question. I was um when I Heard Gernot's discussion. No, you were wondering why don't revenues go up during tax base consolidations And you offered two reasons, but then you had the cave at starting that point was assuming cyclical adjustment is correct And I was thinking what I was thinking about when I heard that was in a sense also the period After the great financial crisis where we saw extremely large revenue shortfalls that you know went much beyond what you would have expected based on normal cyclical parameters and adjustments And so in a sense you can also have a macro environment that works against you. And so you you implement tax hikes You then cyclically adjust, but you still don't see cyclically adjusted revenues going up and maybe There is this thing that really you are implementing that in a very Uh difficult environment and maybe that's one of the reasons which would add if you want to the to the measurement problem problems that both of you have highlighted but I was just wondering how to take that into account in this type of analysis I think that question goes to you. I mean, I didn't want to follow up on that, but I thought Exactly like Christoph highlight that seems like a natural Explanation you have very little discussion of that in the paper. Could it be I would just read So tell me the the main point is so usually when we cyclically adjust we assume a certain elasticity. Let's say Taxes to ggp is what is an elasticity of one? But I think v2 is no longer here, but when he was finance minister of portugal He implemented a very severe austerity program But basically it was in such a hostile macro environment That you you could say the elasticity was much higher And so even though you were hiking taxes, you did not see Tax revenues going up and deficits going down or it came with a considerable lack And that's usually what also the european commission and and colleagues here in the ecb call Revenue windfalls or shortfalls. No, so in some times You see tax taxes growing much faster than you would expect or based on on policies and the opposite when you implement A consolidation program revenues don't react as as you may think on the basis of the typical relationship between macro basis And tax basis. Yeah, so the question is about the use of the elasticity of taxes and spending in Getting rid of these indulgence components of taxes and spending when doing the impulse response And I agree that it's a major issue. And so what we did is was very simple to get we took the Elasticity is computed country by country by the ecd and we took the Average the sample average the result would be the same if we took the country averages But you want to go a step further which is to get the actual elasticity here by here? I think that's essentially what you want to do Which is the ideal thing, but I don't think there is data on that and And it would in a sense it would go against the The very idea of using elasticity is because at that point it's not clear how you distinguish the elasticity the endogenous and exogenous component year by year so it but That's why the first slide that I showed the first impulse response that I showed I think was useful because I agree that there is a a big uncertainty about the The elasticity of tax revenues Which is typically between say 0.8 and 1.5 depending on the tax you're looking at but For government spending particular purchases government consumption investment. There is no built-in elasticity. So the assumption that is close to zero is Is a good assumption. I think and there is that they are very striking. There is a very Very big difference between expenditure base and tax base considerations For transfers the elasticity is negative But it we tend to think that it's small because it concerns mostly unemployment benefits and some Antipoverty measures for revenues. It's big and it's uncertain So I would trust more the results on government spending particular government purchases And there's no doubt on revenues