 Hvala, da me lahko heremonim o moj svetljenju, in dopočne, da so našli, da my se zelo. Zdaj sem glasno, da se izvahno se tega nekaj ljudi, da je to našli, da je to nekaj ljudi. Zelo se, da je vse vse včasne, da je je objev, ko je njega se v moj temošla, ki je delat, in moj privilege, in je tudi, da je tega vse začeljena, neko se, da je tukaj začeli, to pa. A počkaj, da počkaj Matheo, ne sem počkaj načinil, zelo sem očečila v mi, da je malo vozivovati in večanje, vsak razvečnje, zglednje. To počkaj, da ne sem začinila, da počkaj gre za Krotatris, da počkaj Matheo, ne znam, kaj lahko se zelo včal. Bih se, da se pravno, bo se, da se počkaj. Samo všem, da sem pravno. Proste, nekaj smo zelo, da se v centrali banku zelo, da režim je velik faktor in težko monetari politične desetje. V nekaj, nekaj, centrali banku zelo, da je tudi tudi, kaj smo zelo v generi, in da je zelo, da je zelo, da je zelo, da je zelo, da je zelo režim, in even more so, if you think about setting of macro potential policies for instance. And of course, as Matteo said, you would like to do that based on rigorous quantitative analysis and there are various ways of doing it. So what we are doing today is to try to add to the toolkits we use for instance at the Fed and other model and that's actually part of the genesis of this paper together or other joint work or work I have with Francesca Loria, with Danilo Cascaldi Garcia and Pablo Cobaborda, which does something similar with Marco Svici models. But what do we do in this paper? Well, in this paper we investigate the drivers of uncertain can tell risk or future economic conditions, so we look at GDP growth as many of you do in this room and of financial conditions using a stochastic volatility VAR model. And there is a point about forecasting, but there we won't have a lot to add, but the core of the paper is about, or the innovation of the paper is thinking about drivers, so identifying shocks that drive the behavior of uncertainty and risk, splitting in this very simple framework between real shocks and financial shock identifying using zero and sign restrictions. So I usually joke that if you have to fall asleep, do that after watching this picture, I hope this is not going to happen in this room. I guess we are all interested. Anyway, I'm going to give you the paper in one figure and where I'm plotting the one here I had density forecast, which is a regional forecast for GDP growth and credit spreads, and I do that in, and I do that for two periods initially. So the gray density is the forecast of the model in 2006 Q4 for the one year ahead, and that's a period of low volatility, and the black density is the forecast in 2008 Q4, which is a period of high volatility. Max said is essentially what summarizes the mechanism that in this model can generate growth at risk, which is a joint shift of the mean and the volatility of this density. So in normal times you have high growth, low volatility, so you have a pretty buoyant forecast with very little risk and uncertainty, both for GDP and spreads. When you are in bad times, you have high volatility correlated to low economic activity and tight financial conditions, so this density is shift to the left, say for GDP growth, and blow up. So what happens is, first result of the paper, through these mechanisms we can generate growth at risk, so the probability of bad events is disproportionately higher than the probability of good events. But then the second contribution of the paper is to think about what moves around these densities, so we do some conditional forecasting adding shocks, the shocks that we identified in this model. So on the blue density is our density in 2008 adding a two standard deviation macro shock. What you can see, well, actually under the red density is same exercise, we add the two standard deviation financial shock. So what do you see is that on the GDP side, and that's another result, both shocks contribute to shifting the conditional mean and blowing the conditional variance of these densities, while for financial condition is actually, there is some interaction, but it's mostly due to financial shocks. So not only that, but not shown in this picture, we showed that the timing of these shocks and how these shocks play over time differs, so macro shocks are more relevant in the short run, while financial shocks are more relevant in the medium long run. Now, why this matters? And that's maybe, I hope, that's no. Well, it matters because many papers in this literature have used forecasting frameworks to think about the relationship between macro financial conditions and growth at risk, and as we will discuss later in the paper, what we state in this paper is that to be very specific about transmission, you need to add maybe a bit more assumptions. You need to take a stance on the structure of the model that generates these densities, because as we will see, you have macro and financial shifters that can jointly move macro and financial conditions. So if you really want to go to the origin of those and labeling them macro versus financial, probably more structure than some other approaches imposed. On top of it, we do some statistics, or at least we try to think a bit more about what's underlying these densities, and what I discussed so far is the standard mechanism going through the volatility and the level correlation, but those densities integrate all sources of uncertainty, which I think is another important point for applied work, and so in the paper I won't have much time to present it. In the volatility channel, I just presented, but we will also show how estimation uncertainty matters, because in these models, despite using Bayesian shink or whatever, you have lots of uncertainty on the estimated parameters and on the volatility states. And then there is also desire order channel, which again it has to do with the correlation between mean and volatility, and how even when you do these conditional scenarios, shutting down volatility, the effect of shocks on volatility, shutting down estimation uncertainty, you can still have uncertainty and risk moving just because of volatility and the level of the variables covering over your baseline forecast, which we call higher order channel. Now, the contribution of the literature I want to spend two minutes. Max simplified my job a lot, but there is the seminal paper, so I would say in this literature, Adrian Warchenko and Giannone, and then there is more recent work by Plagmon Moller, Lucrezia, Giovanni, all people, you know, very well. And to that literature we contribute by effectively showing that you can replicate and that's something in the paper, the quantiles that you see in ABG using this simple model and you can do it fairly well. And for us that was a practical and a very important point, because, for instance, in the onset of the pandemic when we were discussing growth at risk with Vice-Chair Clarida, one of his concerns was precisely what Max mentioned. We are basing our analysis on a handful of extreme observations. I need to take policy decisions. I want to be absolutely sure you are not giving me results based on three observations. These and the Marko switching paper kind of really got, gained traction internally. But we also build on the shoulder of giants. I mean, we have the giant Max here and this model we estimate is nothing new. Max has estimated many of these models. What we show related to his Max work is the implications of this model on the full densities as many of the papers focus on how volatility shocks shift the conditional mean of the distributions. And then linear vars. Great tool. I don't need to do this in this room, but we get us why can't we just look at an uncertainty shock in a linear var. Well, the reason is surprise, surprise, because you will have probably densities that look more similar to the gray density shifting around without any statistical notion of uncertainty and risk. And then we get to the storytelling. What we uncover, it's kind of plausible, because what we uncover is macro and financial shifters both contribute to growth at risk, let's say, which is a feature of many DSG structural models. And probably the nonlinearities arise both on the macro and the financial side. That's kind of underlying this dual, the de nexus. And you can think of financial accelerator mechanism for particular work, but also effective lower bound on, say, the interest rate as behind some of those results. So I'm gonna skip the plan from the talk because I'm already talking too much. Super simple model. The model is from a paper by Harun Mumtaz. And it's a model where what we really wanted was to have a rich correlation structure between the mean and the volatility of our macro and financial variables. Level equation is a var specification where there are two key ingredients. One is you have this volatility processes h that enter linearly with lugs so you have this delayed effect of volatility on the level of the variables and then you have this contemporaneous multiplicative term, which is the stochastic volatility. So our level reduced form residuals are ex-CV time varying volatility. Volatility is endogenous in this model as it responds to is an autoregressive, but there are the level terms entering in the equations and then there is their own shifters. Now, when you estimate these models, you typically find that there is correlation between the level, the residuals in the level and the volatility equation. What's nice about Harun's specification is that it allows for a covariance between the two that you can estimate from the data. So that even more gives even more kick to this joint mean and volatility correlation. I'm not going to talk more about that. The model specification is what it is. Now, the data is we are very parsimonious, so we are not in the big model league and we do it because we really wanted to get at the core of the structural identification. Why do we use these two variables? Because in the working paper version you will find real time estimates. So here you will find full sample estimates. But in the original paper we used the Philadelphia for real time data and we started in 1970s and we have real time measures of uncertainty risk. Whatever trust us on the posterior algorithm and we did a decent job at estimating this model. Very quickly what I discussed in the conditional distribution that we produce factor in all sources of uncertainty underlying this model. So there is the future realization of the shocks there is uncertainty about the state's volatilities and on the parameter of the models. And this is a point that relates a bit to the work of Blackboard, because they claim there is so much uncertainty around some of these dimensions that is very hard to tell whether the conditional mean and density and even no, the conditional mean they say we can identify the shift. It's more higher order moments are hard to identify. What we actually claim is that that's a feature. Yes, you're right. There is a lot of uncertainty. But when you integrate it over it you can still make meaningful statement and you can still have meaningful estimates of this nonlinear interaction. But we agree with them. Not much time varying skewness or kurtosis. That's not something we can detect. So I showed you the conditional density. So this is just we characterize some moments from these conditional densities. But we in the paper I'm gonna show and what I'm gonna show we plot impulse responses of these densities. And the only point I want to clarify here is that for instance when we construct an impulse response for uncertainty, we are again constructing an uncertainty measure for say a baseline density. So the square root of the variance of that density say the black line before at a given horizon F. But and then we are constructing a counterfactual density where we add shocks. For instance we add one standard deviation or two standard deviation micro shock. And then we take the difference in the moments. So that's what those the impulse responses to uncertainty to the downside risk. So this shortfall measure and the long rise, the upside risk are gonna be and as you can see we integrate over the parameters and the states. So sometimes people are shocked that we don't have a notion of like we don't have confidence bands around our impulse response. Those are integrated. They are integrated in. And we show now some alternatives where you the dispersion of these objects when you fix the parameters and the states. Identification. So that's gonna be that's kind of where these models come handy because again we can apply our arsenal of identification strategy in a relatively straightforward way. There are some couple of things that one needs to be careful about but by far enlarge is a straightforward use of zero and sign restrictions. So rewrite the model in structural form and without loss of generality you can put GDP on the left on the first equation and I'm gonna call that's gonna be my structural call the first shock a macro shock and the second shock a financial shock. So let's start with those two. So we want to strike a balance between we don't know really how these shocks can be identified. There is so much differences in what models tells us that we wanted to characterize a set of models so we're gonna do set identification but we wanted to make sure we can label the shocks. So what we do is that we use sign restrictions where possible. For instance in mediating the relationship between spreads and real activity so we argued that the endogenous and the intuition of identification is always the same. GDP or spreads can move as a response to cyclical variation so there is an endogenous response of these variables and there are exogenous shifter. So endogenously we think that all else equals spreads goes up, go up GDP goes down. That's one assumption and for spreads same thing GDP goes up, spreads go down and that's not enough of course to tell apart the shocks so we rely on bounds and we don't know exactly how to sign restrict the relationship between GDP volatility and the level of the variables because there are some theories where there is uncertainty as GDP goes down but there is also some option value of waiting activity might go up. So what we assume is there can be a response the response can actually be pretty sizable I'm gonna explain it but needs to be bounded cannot be too large so these bounds we said they are using numbers which now I'm thinking is a terrible way of showing this but what is the idea underlying these bounds so is the following on average macro volatility goes up by one standard deviation so about 10% GDP can move by up to one third of the standard deviation of the reduced form residual so think the reduced form residual in this model for GDP growth on average is 70 basis point that's the unexplained component we allow macro volatility to explain one third of that which is a pretty sizable which is a pretty sizable number and with spreads we allow even more the reduced form spread residual as a volatility on average of 20 basis point we allow the macro volatility to explain up to 10 basis point of that surprise so up to 50% can be explained by volatility more than that we don't allow that macro volatility can entirely drive for example the surprises in spreads we just don't admit it in our set so we try to be loose but impose some structure on what we think is plausible last point on identification then I need to move on why do we have these zeros let me give you an intuition in these models you have a pretty large relation between financial variables and their volatility in fact some of these financial variables have captured risk so on the one hand we want to have these densities exhibiting these profiles with moving tails uncertainty on the other is very hard to tell apart what are shifted to the level or the volatility of these variables because pretty much as we actually shown in the paper there is one shifter which is a financial shifter that moves the level and the volatility of the spreads so we take two extreme assumptions in our baseline we assume that shocks like financial volatility show anyway the effect of financial shocks or financial conditions on activity is entirely mediated by spreads and volatility of spreads we set to zero on impact then the dynamics of the model are free to do whatever they want we also take the opposite sense so we rewrite this system as an alternative where our financial indicator is the volatility of the spreads and we don't allow the spreads to do much on impact you get the same impulse response functions very similar and that's because these two variables are highly correlated in fact in many models you don't even have financial volatility you just have spreads like Ludviksson, Ma, sorry and then bottom line this is the core we have two other shocks now we don't care much about these shocks why because we identify them and there is a full endogenous movement so in our model as I discussed volatility matters and certainty matters however once you characterize a macro and a financial shifter the other two shifters don't generate much business cycle for movements they move our volatilities they move a tiny bit GDP growth, a tiny bit spreads but they don't it's kind of almost like a residual variation so we are not gonna work much with those two shifters but and I want to be very clear before using my last five minutes to discuss the results that doesn't mean we don't believe or care about volatility shocks uncertainty we just use a different labeling system we use financial shock to denote a shock that generates co-movements in volatility and spreads like the GFC to us is a financial shock cannot call it level or I cannot call it volatility because these two variables move together almost all the time going to the results ugly picture I'm sorry this picture gave us the biggest headache because these are the unobserved volatility states for GDP and spreads what you can see they look moving over the cycle per seklikality of these processes however while the spread volatility is very precisely estimated for the reason I told you it comes with spreads so there is a lot of uncertainty around the estimation of the GDP growth volatility process now what is the thing to be keeping mind these are point wise credible sets so yes they look large however I don't care as much about the point wise credible posterior sets what they care about is is uncertainty moving interseklikali or not so what did we do to prove that the answer from this model is resounding yes despite this much uncertainty we constructed the posterior probability of uncertainty of volatility at time t being larger than in the past we did it one year ago you can take the average over whatever you can do it whatever the red shaded areas represent periods or quarters when that probability is above 75 percent every time you are approaching a recession or you are entering a recession this model is gonna tell you you have higher volatility with extremely high probability and that to us was okay we are in business wow I thought you would give me the five I didn't see it I'm already at three we have maybe three pictures so let's see first I'm gonna show you traditional impulse response functions to just discuss a couple of things about the non-linear mechanism in this model so here top panel I'm plotting the effects of one or two standard deviations no maybe one macro shock and the bottom is a financial shock the red IRFs are in periods of high volatility so this is 2008 the blue are periods of low volatility so what you can see is this non-linearity kicking in in the model of course the effects of high volatility during periods of high volatility of this shock is disproportionately larger the second point is both shocks generate business cycle co-movement so spreads go up, GDP goes down volatility move up as a bonus which we didn't exploit too much so we even have some volatility reversals one point that is underappreciated is what this very wide credible set tell us is we have this co-movement the co-movement is there we have a lot of uncertainty about the quantitative implications of this model but if you factor the team that contributes to your measure of uncertainty and that's important because what we were wondering is GDP growth volatility doesn't move much how can it be that our GDP density has quite a bit of bad realizations in the tails well that's because for macro but especially financial shocks you have a lot of dispersion in the potential outcomes because for different parameters and states you can have GDP dropping dramatically or maybe not moving as much and this gives you a sense of that statistical uncertainty and plotting the size of the credible sets over the horizon and what's interesting is that the dispersion say of the effects on GDP growth of these shocks is almost as large as the level effects that these shocks have so this implies that just being uncertainty and being dispersed about how these shocks evolve can generate substantial uncertainty that's something we need to tell our policy makers then they can take a stance if they love one parameterization they are going to pick the one they love but we shouldn't hide that this is the case and that contributes to our uncertainty or maybe we shrink like crazy and we get rid of it but right now that's how we are going now I have to show you in my last minutes at least one picture on uncertainty so these are the impulse responses of those shocks on our measures of uncertainty and tail risk and the importance of the densities I showed you in the paper in one figure when you have a macro in the financial shock uncertainty around GDP growth goes up a lot in the short run and the macro shock actually matters a lot is as important as the financial shock as you go as you advance your horizon the importance of macro shocks becomes relatively smaller but at longer horizon macro shocks matter less than half is like the responses less than half of the response to an uncertainty shock and that's the standard persistence of the effects of financial shocks kicking in same with spreads again is financial shocks that play the lion's share but macro shock matter because there is this joint transmission of shocks all our shock transmits through macro and then last point I want to make you dislocate the tails of this density and when you move the tails they don't move in lock step and that you can clearly see when looking at the response of the tails of the GDP growth distribution to a financial shock the shortfall so the left tail of the density can move up to two percentage points to the left the right tail of the density barely moves and especially at longer horizons so the mechanism of shifting mean and blowing volatility does really well at dislocating the tails in very different patterns and I think that's gonna be it one other bonus from this model is that and then I'm gonna cross that shocks have intuitively asymmetric effects when you have a good like a bad shock if you think about it you have spreads going up, volatility goes up you are more likely at that point to draw terrible financial condition like spreads at 500 because you are blowing the volatility and spreads are already high when you have good shocks you reduce spreads you reduce the volatility so the effect of those shock I cannot click otherwise you will clearly see of the picture all uncertainty in volatility because and that explains also this asymmetry of uncertainty and risk measures over the cycle because when you recover uncertainty doesn't drop as quickly as when you went up and that's precisely the negative correlation between the mean and the volatility that's it before I get the zero or the warning or whatever let me conclude by this is a simple model but we did it on purpose this issue of thinking about the transmission of shocks to uncertainty and risk and we wanted to do it right you can extend it to larger models that's something we are also thinking but I think the bottom line is both it seems that both macro and financial shocks matter when thinking about uncertainty risk dynamics thank you and the discussion is Danilo Levalão thank you very much to the organizers for inviting me to discuss this paper so Asdario sorry just mentioned so the goal of this paper is trying to assess the drivers of uncertainty both uncertainty and the risk of future GDP growth in doing so the authors employ three main ingredients they have a base and VR model with custom volatility and also with feedback between the level of the variables and the volatility the second ingredient is the identification of structural shocks through a set of sign restrictions exclusion and also magnitude restrictions and the third ingredient is basically the computation of the effect of the shocks on both on tail risks and uncertainty so I think that the key result in this paper is basically the last picture that Danilo show which makes reference to how the tails of GDP growth distribution react to macroeconomic and financial shocks in which we can see the asymmetric behavior in this response so in the left panel so what we can see is that the response of the left tail of GDP growth which is given by this solid line is stronger than the response on the right tail of the distribution of GDP growth which is the the noted as a long rise now something that is important to emphasize is that these risks are computed from conditional densities of GDP now in order to understand a little bit more the nature of these results so let me show you a few pictures that basically kind of illustrate the type of densities that we will obtain with these type of models so this is just with artificial data simulated data and we can see like how these densities will evolve four steps ahead in the forecast horizon so we can see the median of the distribution will shift to the left in this case as the horizon increases but also the standard deviation or the dissipation of the distribution will also increase now if we do some sort of aggregation or like in average or some manipulation with the distributions at the end of the day so if we aggregate all the information containing these four distributions we end up with this other distribution basically a skew distribution left skew distribution now in this skew distribution we can compute the shortfall and the corresponding loan rise associated with the 5 and the 95 percentile so this will constitute our baseline forecast now what Darian co-authors do is constructing a counterfactual scenario in which there is a shock so there is a shock that hits the system and this moves also the predictive densities in this case these are adverse shocks that is basically moving all the four distributions towards the left and also increasing the volatility and then we can also do the same exercise so we can aggregate all these distributions in red and we will get something like this and we can also compute the corresponding shortfall and the corresponding loan rise now the difference between this blue line over here vertical line and this vertical red line over here in the bottom chart is what constitutes the effect that a shock has on the left tail which is basically what Darian co-authors call the impulse response on the shortfall and similarly the difference between this vertical line over here and this red line over here is what constitutes the effect of a shock on the right tail of the distribution which is also the impulse response in the loan rise now here is where we can see the asymmetries that Darian show in his last picture now this is with simulated data and this takes me to my first remark which is where exactly asymmetries come from now what will happen if we shut down this feedback between the level of volatility involved in the model which seems very key so what I did is to estimate a standard based on VR model with stochastic volatility without any feedback whatsoever with information on GDP and spreads up to 2008 Q4 and at that point so what I'm gonna do is to produce four periods ahead forecast and replicate these sort of graphs but remember that this is with artificial data but in this case it's gonna be with true data GDP and spreads as well so what we see is here we have the baseline distributions in blue the original distributions and the aggregated one and in the bottom charts we have the same but for the counterfactual that is when we have a shock now if we compute if we focus only on the one quarter ahead forecast density and we compute these impulse responses by following the procedure that I just described so what we see is that the responses are very similar so there is not so much really asymmetry in the response of the tails obviously in this model we are not allowing for this feedback between the volatility and the level however what happens if we construct the same impulse responses but based on the aggregated distribution in this case the picture changes considerably because we do get some asymmetries in the distributions now notice that these asymmetries that we find are not really a product of any feedback between the volatility and the level it's just a simple issue associated with the aggregation of the distribution now so this takes me to the point that the level of volatility feedback is crucial to generate asymmetric tails that has shown but this also is the aggregating procedure so I think it's very important to understand what is really driving these asymmetries that we observe in the impulse responses ok my second question basically that you already commented that there is another work in paper in which they provide real time estimates because this will be very very useful for policy makers so I'm gonna go straight to my third comment which is about the COVID period so the estimations in the paper stop in 2019 Q4 so then I was wondering how would this sort of models with this feedback between volatility level would perform when facing the COVID period so what I did is just to take the original model of moon tasks in 2018 and estimated with data again on GDP in a measure of a spread with two samples and estimate in particular what I'm showing here is the estimated volatility of GDP growth the blue line makes reference to this volatility estimated with sample that stops in 2019 Q4 and the red line is the same volatility but estimated with a sample that stops in 2022 Q2 so as we can see is that the model produces reasonable revisions in the growth volatility once we incorporated the COVID period so I think this is a pretty nice feature of this sort of models with this feedback between the volatility level that perhaps should be even emphasized in the paper in particular in light of the recent turbo lenses in financial markets so I think it will be very very much welcome to updated estimates with the information up to now now I'm gonna finish with this a few additional remark which are associated with some of the results from the paper so the first one is associated with the structural part so one of the results is that the effect of financial shocks are last longer than the effect of macroeconomic shocks so I'm wondering at which extent this result might be induced in the financial restrictions that the authors assume that the financial volatility is not allowed to affect contemporaneously all the variables in the system but macroeconomic volatility is allowed to affect contemporaneously all the variables in the system so there is some lag relationship which I don't know up to which extent if it might be influencing a little bit this result so in the second comment macroeconomic and financial shocks lead to an increasing uncertainty so here I'm just wondering at which extent this might be associated with the way in which the authors construct these impulse responses because on the one hand the authors have two distributions so one of them is without the shocks the other is with the shocks and then subtracting the standard deviation of the two of them but in the second one an additional source of randomness there so perhaps this is increasing the standard deviation associated to this second distribution so I'm just wondering if this is the case and the last one is about the result that claims that the effects of shocks are stronger in periods of high volatility so when we employ these sort of VR models with time variation in order to assess the effect of given shocks so there are basically three things that may be going on so the first one is that there is actually a change in the transmission mechanism from the shocks to the variables or perhaps there is just a change in the size of the shocks but without any change in the transmission or perhaps the two things are happening there is a change in the transmission and also a change in the size of the shocks so I'm just wondering whether the model can tell perhaps a little bit about this about this which of these three scenarios are we facing but anyway, I just want to finish the discussion by saying again to Dario that I have enjoyed a lot discussing this paper, I think it's a very very interesting paper, it has a lot of features I would like to see the real time results and I especially think as Matteo Clay said in his introductory remarks so this is extremely useful for policymakers, so thank you very much so let's take also some questions from the audience and then I will give the floor to Dario to reply Hello Hi Dario, nice paper, nice discussion, Danilo one question is like I was nicely surprised to see that the great moderation still holds in your ok then something that I was wondering like and actually something that happened with a lot of these models that a lot of the commoments come from the fact that the commoments are different before and after the great moderation and I was wondering if there is a possibility in the same way that we control for the COVID and we consider this special thing I don't know if you consider the possibility of analyzing different parts or like taking some kind of like control for that Thank you Dario for the great presentation coming from the uncertainty literature I'm a bit surprised about the results that you have, especially if you put in what we learned from Ludwigson and Anma that you have financial uncertainty that is actually the exogenous component when you think about these different measures of uncertainty so I would like to hear your take on that and the second question is whether you have projected that these macro and financial shocks on TFP, monetary policy fiscal shocks, like a check of whether these are exogenous things or they are moving with other things that we think are exogenous So let me then give the floor to Dario to answer the question I go first So thank you so much this was a great discussion I wish all discussions were so thoughtful as yours and all referees were as nice as So quick replies, I really like your exercise, so the level volatility feedback matters but you are right that the aggregation also matters we construct these aggregates because we wanted them to compare we have it in the paper to the measures by Adrian Boyerchenko and Giannone which look at when they run their local projection look at growth between E and T plus H so they kind of construct that same counterpart so in the paper we constructed the counterpart but you are right that that's something we should apportion at least how much in the short run you still see that you have some asymmetry but I think it's an excellent point what about covid we had it in the paper so actually if you look at the working paper this version probably is very well suited for us central bankers it didn't fly extremely well with academics because we did both the real time and the covid exercise which we then wanted to spin off into a separate note but we never had the time so you are right the model doesn't do bad adding covid data the problem is that the forecast you get are completely nonsense so what we did to do that is that we run a slightly different so we re-centred the forecast around SPF and then we said if we re-centred the forecast around SPF what are measures of uncertainty and risk that will be associated just because the model has too much persistence so you start the forecast in 2021 you know q1, q2, q3 you first get this harmageddon which is going to last forever then you flip that because you have this mass so there might be other ways we did that it works you are right but those extra steps were like the referees were like what the heck are you doing there it's like not fitting with the rest of the paper but that's a very a point very well taken on the grain moderation I took points so one is in the real time exercise one thing that you notice is that it will take us 5, 6, 7 years longer to pick up the grain moderation then in the uncertainty estimates we produce here so here it's almost like sad in 85, 86 when you do it in real time the model for 5, 6 years doesn't see it so uncertainty remains much higher and then it's more gradually declines I think there are ways of thinking about these long run trends I never got too much into it but I think some paper by Ivan Petrel and Quotor so I don't know if some of you are here but low frequency and high frequency movements in uncertainty and what are the implications for growth at risk and I saw it presented a few years ago it was great so I would just defer to that paper and I need to read it more carefully and on Pablo's question I think our results are actually very aligned to LMN I don't know I go by acronym but Ludwigs on Ma and then it's just that financial volatility is effectively financial conditions they don't have a measure of spreads they don't have a measure of stock returns in the paper is a three question var with two volatility measures and IP if I don't remember so when you look at their impulse responses actually they align pretty well for the level effects because it's a linear var so that's the thing but the level effects align pretty well with our impulse responses on the level variables just the labeling is they call it financial volatility we call it financial shock but if you compare those two impulse and again to us is intuitive like it's risk and you know these variables are themselves indicator of risk I don't know what to call of like do you tell me that their shock is different from the Geekritzagrasek accept bomb premium shock I don't believe it because I know I mean it's the same impulse responses again these are like disturbances originated in the financial sector that generate this big shift in how tight are financial condition and measures of risk and volatility so that's how we are interpreting how we are placing ourselves, thank you thank you Dario I understand also your line community seems to be a bit more quiet than the in presence so I guess we close the session here so big applause to all the speakers