 The main question in the paper is how are institutional potential output estimates revisely responds to a shock and what seems to be disturbing in that evidence is that in presence of a shock with just temporary effect on GDP, even in the long run we see basically an important revision in potential output, even in the longer on when, even in presence of information on asymmetries people estimate in potential output should have understood that the shock is only a temporary one ne če ne izvajte potencijalne izvajte. Ism je tudi posil všim? Slepa, da je pomečena, še prist, in je tudi pravdiv. Tudi, da je vse zelo, je vsezal tečno zelo v vidjejo v VAR, izvajte v šarj, identifikacije, skimovati, z zelo v zelo kaj posil, iz kaj vsezal, in počem neslipa potencijalne izvajte z vsezavljenih z pristi, da ne vsezal, Or, its revision do not overreact to temporary shocks. And when you look at the policy implications of their estimations, you find out that if you take their analysis, then the U.S. of Pogap in the last decade is probably much larger than we thought. And I'm very sympathetic to this message with my colleague, Mario Kiarosinski, who was working on euro area data in a very different paper, but we come up with two relatively similar implications on the euro area potential up put and the up put gap. In particular, we find that its best potential output in the euro area after the crisis is best characterized by a relatively temporary slow down, but then return quite quick to pre-crisis growth rates. And then also to a much larger up put gap than institutional estimate. So you see a relatively similar result even based on very different methodologies. Now, I think that this paper has a very nice byproduct, so in particular this relatively simple and powerful, time-serious method to cross-check the estimate to potential up put. I think it's something that we should consider for the toolbox for the briefing monetary policy in central banks, in particular to help the assessment of the risks to more judgmental and sophisticated estimate to potential up put. Of course, these estimates and those that we have in the paper with Mario Kiaroski is also a lot of question about the policy implications, particularly about whether we should have done or we should do more policy stabilization policy than we thought. And the other one is what is the level of the natural rate in the recent decade if we take that potential output didn't strongly slow down. In particular, this would be relevant for discussion on the new normal. Is the natural rate so low as we thought or it's at the levels that we had before the crisis? Apart from this very general consideration, I have actually two further, a little bit more specific comments. The first one related on the issues, further issues for the business of estimating potential up put that can come from the advent of an economy more based on intangible goods. And in the second one what I want to do is to test a little bit the proposal of using Bevert VARs and in general the Blanchard Equal Methodology or EuroArea data with the aim of highlighting some of the possible challenges that there would be if we want to use these types of methods in the actual briefing at central banks. So, let me start very quickly with this comment on tangible intangible. We know investment can be into type of goods, so tangible goods is those you can touch, the usual investment we have in mind. And in tangible investment, which is said things you cannot really touch, so computerized information, innovative properties, economic competencies. Now, people working in this literature know that a recent finding is that over time in the share of intangible goods into GDP is growing while the one tangible investment is decaying and actually you see that in the recent decade intangible investment as a share of GDP seems to be larger than tangible investment. Why is that relevant? Well, because this actually poses another measurement challenge beyond those that Olivier already mentioned for estimates of potential output and for many other things. We know in fact that a large part of intangible goods and intangible investment escapes the national accounts. So, we might be much more uncertain about the potential output that we actually think on the basis of the data that we have. Of course, when you look actually at the estimates of intangible investment, it looks like at least so far that there is much more of a level shift rather than a change in the dynamics of investment due to this intangible investment. So, it's possible that there will not be such a big issue for our models but it is something that we have to take into account when we think about how uncertain we are about potential output, the apple gap and so on. Another thing which I think is relevant to take into account for the issues discussed by Olivier is that an economy that is strongly based on intangible goods could actually become more prone to decaying slope into, so decaying slope into the development of potential output. Particular because we know that for some of the characteristics of intangible investment, for example, scalability, implies that we might go and we already seen that to an economy with a few mega firms and an army of laggards. And these army of laggards might possibly imply important under investment. So, endemic under investment and given the relationship of investment in TFP growth, we would basically be heading towards endemic low TFP growth. So, I guess that we shouldn't be complacent about the results we have found both for the URI and for the US over relatively resilient slope of potential output because we might see in the future something different due to this advent of more intangible, more relevance of intangible goods in the economy. Let me then dive to my last point. I was really strongly inspired by the time series logic in the paper of Olivier and Quotors. Particular the main example which is based on this Blanchard and Qua methodology I think it's a simple and powerful idea and it's very tempting to use it in real time policy making although, of course, to fulfill so, in order to be used and to be accepted in our toolbox, it starts to fulfill several criteria that I want to mention and I want to discuss. I will focus mainly on one aspect in particular because this is not characterizing only the Blanchard and Qua application but also some of the other applications proposals that Olivier and Quotors presented. Particular what I want to ask here is does this method, this Blanchard and Qua method and the application in the paper is able to properly disentangle shocks and propagation. So let me tell you what I mean with that. So when you run this exercise based on VARs and implicit assumption before you start identifying shock is that your VAR is actually able to disentangle the propagation so if you want the red part up there in the equation from the shocks. So the red, the green part there. So you would expect that the forecast error of the VAR are only a linear combination of the structural shock and then you can run your identification exercise using the identification assumption that you believe in. This boils down to have that YT so the observable you put in the model the vector of observable is large enough to capture all the relationship of GDP and unemployment with the rest of the economy. In other words you do not have to omit any variable that would grandeur cause these two variables. And you know there are ways of texting that Mark and Bartoschier in first row they have a very nice paper on how to do that in a Bayesian context. Generally I would say that this bi-variate VAR from what we know about forecasting in the US and in the Euro area is too small to identify all the shocks. So most likely there will be omitted information. However let's remember what this exercise really is. They don't want to identify all the shocks in the VAR. They only want to identify really those that have a permanent effect on the economy. So it could still be that to this relatively smaller task this VAR is able to accomplish even in presence of omitted information. Of course this is an empirical question. So what I want to do is just to give you an example of what I'm talking about using Euro area data. What I will do I will compare a small VAR to a larger one where I put some information that would possibly be taught to grandeur cause the two variable here and I want to see if there are big differences in the outcomes of these two VARs to find out whether the omitted information is indeed relevant. So what I do here is just to estimate beside the two variable VAR Olivier Tal have estimated a five variable VAR where I add the real investment consumer confidence and capacity utilization. I run the VAR levels because I suspect that there are long run relationship I do not want to wash out by differencing here. I use quarterly data from 1995 to 2017 and not 2007 sorry use Bayesian estimation and I estimate so I identify the shocks similarly to what they do in the paper although with some small twist that is probably not appropriate to discuss here. Let me try to give you some results. So in the top panel here I'm showing you the shocks and the red line is the shocks in the bivarate VAR while the dash the black line is the shocks in the larger VAR you see that they are not that different but there are some important differences in particular around the period of the crisis where the bivarate VAR goes much more down and much more up and stay more positive for a longer period of time so we could you know we would think that that model adds some deficiencies into capturing some of the dynamics that were probably captured by some of the endogenous variable I have added in the five variable VAR and I'm looking at the impulse responses there to this shock and again the dash the black line are the impulse responses to the long run shock in the five variable VAR 16 to 84 quantile in the median and the red line is the median of the bivarate VAR so the density of the impulse responses are actually partly overlapping but the median of the bivarate VAR is sensibly higher than the median of the five variable VAR on impact twice as much the response and generally staying on the upper edge of the response so it seems that this five variable VAR does add information which could be relevant for the problem attend that is I remind you estimate the supply shock and throw them into this transmission mechanism to come up with an estimate to potential output so let's put it all together and let's look at which kind of estimate you would have for the upper gap in the two models now the level of this estimate doesn't mean anything because what I have done is buy the properties of the exercise and normalize them to zero in 1996 so they only have a sense of where would the upper gap be in a certain period compared to 1996 but the two are actually comparable because they did the same normalization what you can see is that the two models again do not differ that much in terms of peaks and troughs but in very important parts of where we would like to know where is the level of the upper gap they do differ for example in the boom period the upper gap in the euro area was sensibly higher in the red line and very recently we see that instead the red line would probably give us a more dire perspective than the dashed red line but the difference in the last part is consistently around 2 percent now it is also fair to say that if you put uncertainty around these estimates these differences are not enormous okay maybe there is a significant difference there in the peak of the boom and very briefly in the last part of the sample but by a large, statistically these differences are not too big however, we know that if you are a policymaker you have to talk about the level of the upper gap zero is very different from minus two also in terms of communication so let me then say what are some very quickly what are some other dimension in which we would want to test the proposal of Olivier encotors if we want to use it in our tool but of course there are other possibilities in terms of which variable to add there is a way, a reason to think that we should be much more careful about the domestic component in the VAR they mention that in the paper as well so in their model, in my model here is just exogenous but you might want to model it for example as a function of slow moving important elements like demographics and so on and so forth there is a question of how much uncertainty there are in these estimates there is a question of robustness to data revision and so on and so forth so just to say that this is a great proposal I think we should think a little bit more carefully to how to do to include it in our toolbox so let me conclude now so I think this is a very nice paper it gave me a very fresh perspective on an old but still relevant problem I think that I agree with them that maybe potential output is not growing so slowly after all and in my comments I just tried to twilight some of the issues around this work or if you want elaborate on some of the further question that are stimulated by this nice paper thank you okay now is time for round of questions before we also turn back to Olivier to react to the discussion we have two mics running around and Mark has the first question if someone could give them a mic yeah it's coming from the other side Mark I was just gonna say that I think it seems quite reasonable up until this past year argue that outputs been below potential just on the behavior of inflation because in the US inflation have been running at one percent to one and a half percent so that's consistent I think with your argument but in this past year my impression is there's been a pickup in inflation to around target two percent or slightly above and you're seeing wage growth real wage growth start to pick up as well so it seems like to me what you kind of want is you have these sizable output gaps up until this past year but then you'd kind of like them to narrow at least based on the inflation data next question over there yeah Frank to the right of you two questions one question is I was wondering to what extent these results are driven by the great recession and the double dip recession in the case of the euro area because there of course don't mean the financial crisis was a very important element and you could argue that when financial intermediation isn't working that that has an impact on sort of misallocation of resources and TFP if we do a decomposition of this fall in potential growth for the euro area for example in the recession then it's mostly coming from the TFP growth again that's not any evidence but it's just a sort of an observation the second question I have is so I mean basically have kind of three theories out there not to explain the cyclicality of potential growth or the level of growth one is kind of mismazement which I think can be statistical mismazement or can be the fact that we forgot or we didn't allow the slope of the Phillips curve to flatten because for example in a low inflation environment there is more evidence of downward nominal weight rigidity and so on so forth so that's one sort of set of theories the other one is the other extreme where this is indeed the sort of the legacy of let's say the recession on potential on the working of the supply side of the economy and in between and you mentioned that there is sort of the hysteresis kind of hypothesis I wasn't sure where your analysis speaks to that hypothesis and so again another question Bernard in the back yes somewhat going in the same direction it seems to be a healthy reminder of the kind of work that Ofanidis did on the pitfalls of the real time measurement estimates so I would draw the opposite policy conclusion that Nikkele has done that there's more scope for stabilization policy maybe there's less scope and the two points I want to make is one are your results robust to the starting point so isn't there as much overestimation of potential output before the crisis during the new economy boom during the credit boom bus cycle then there's now a case for seeing an underestimation so on average maybe we actually in level terms we are not underestimating now if the overestimation pre-crisis was high enough so that's one question on robustness and the second point what is your take on the recent GDP revision for the US in the benchmark revision so we actually have uncertainty about actual GDP not only about potential GDP we've seen actual GDP revised up now by one percent so the gap you're pointing to is being closed not just by revising the potential output but also by the real output the actual output thank you okay I will come back Arresta but before at least I forget the questions maybe we go back to Olivier sure thank you so thank you Michela for your comments I think the idea of using a larger var is a fine one we don't have any particular attachment to the specifics of the Blanchard and Qua approach for us this was just a particularly tractable way of trying to separate supply and demand shocks in real time as Michela emphasized the Blanchard and Qua approach does not allow you to separate out the different kinds of supply and demand shocks and so in principle it will work only to the extent that different supply and demand shocks give you pretty similar dynamics you know in practice it seemed to work fine in terms of what we were doing but certainly there's I think there's scope for considering larger scale vars one issue with integrating consumer confidence into these vars is that previous work has found that consumer confidence seems to capture a lot of news about future productivity and so this is going to be a little tricky to integrate when you're thinking about estimates of potential GDP because if you're getting news about a future increase in productivity which would show that should tell you that you know in the future potential GDP will be higher potential GDP today has not increased and so if you're integrating this consumer confidence into your var then you're likely to overestimate by how much potential GDP is changing today because you're going to be incorporating this news about future productivity into that but certainly I think there's scope for considering larger scale vars in principle you know that you should be able to combine the information about consumption and inflation et cetera into a single framework which I think would be much more useful in response to Mark's point about inflation over the last year so in the context of the Phillips curve we're using what matters in terms of identifying the gap is the difference between inflation and the measure of inflation expectations that you're using in that Phillips curve the inflation expectations we're using are still higher than actual inflation so we would still have a gap of the same sign it would be the case that that gap is smaller now than it was as inflation kind of converges higher to those expectations so in the paper we stop in 2017 so that's why you don't see that closing happening but certainly would have narrowed somewhat over the last year if we extended it it wouldn't have closed again because those expectations are still significantly higher than actual inflation so in response to the point about hysteresis so we view our interpretation of the data as complementary in terms of the policy implications as the hysteresis view so the hysteresis view is you've had these demand side shocks which have brought down the potential level of GDP through whatever hysteresis mechanisms we think would be important but if that's the case then demand side policies can reverse those changes and so you should be doing more of these demand side policies to reverse the declines of potential GDP our view is that potential GDP hasn't fallen so much and there remains a large output gap which you should be closing through demand side policies how do you separate between those our view is there's not there's not really any empirical evidence for the hysteresis view there's a long literature on for example the effects of monetary policy on the economy you don't find in any of those papers hardly results that suggest that monetary policy shocks have long lived effects on GDP which is what you would need for the hysteresis view to be important of course in the long run the fact that you can't reject that the effect is zero doesn't mean that the effect is literally zero so maybe it's just a question of precision about the long run effects so when we do the Blanchard and Qua we're implicitly assuming that the hysteresis view is not true but when you do the the Cochran approach the Cochran approach is not taking a stand on hysteresis all it's saying is you can identify changes and potential from what's happening to consumption based on whether people are interpreting those changes in income as being permanent or transitory so imagine a world where you have hysteresis then demand and supply shocks can have permanent effects on GDP that's going to be getting picked up by consumption in this world and so when you do the Cochran var those declines in permanent GDP in the potential GDP that we're identifying could be coming from either supply or demand side shocks the demand part can happen if you have hysteresis in that case we're not ruling out hysteresis but qualitatively it still gives us the same picture which is there's been some decline in potential but not nearly as big as what the CBO would have us would have us think the point about linking this to Orphanides we agree I mean the interpretation we have is kind of the flip side of what Orphanides said about the 1970s and so essentially right now I think we're at risk of understimulating the economy because of mismeasurement of the gap and then in terms of where we overestimating the gap pre crisis that's certainly an issue when you're doing the var methodologies because they're essentially we're imposing a normalization as as Michele said about what was some initial level it's not so much an issue when you do the Phillips curve approach because there you can actually make a statement about the size of the gap and in that case it's giving us something something similar but certainly when you do the var methodologies you do have to worry about what is the starting point that you're assuming Oreste he left okay so was not so urgent any other questions very good so when policymakers get presented these estimates from the various models present us are not always as lucky as as you seem to be where all your estimates are above the official estimate but often time they kind of cross all these lines or they spend pretty much the entire chart and then it's everyone's guess which one is the most relevant one but what I take away is that we should think very hard about whether we are facing series of demand shocks or supply shocks and this is also in the forecasting side of the ECB where put a lot of emphasis not just relying on the models but really thinking about the narrative that goes into these models all right thanks very much for this presentation so we move on to the second paper Emmanuel will present this