 Good afternoon everybody. I'm calling for the USO for me. It's actually still morning. I'm very happy to introduce this next session here with really two very interesting papers on very, very topical issues I would say. One is dealing with trust and monetary policy and the other one with managing the monetary policy normalization. These are topics that are very much into the minds of central bankers in general, but especially in the current environment. And the first speaker who already has a kind of switch on his camera. And I'm very happy to introduce because he was my PhD supervisor, so I know him quite well. So I'm very happy to see him, even if it's only remotely. It's Paul de Graua. He's currently at the London School of Economics and he's presenting the paper on trust and monetary policy that he has written together with Jomi Yeh, also from the London School of Economics. So Paul, please go ahead. You have half an hour for your presentation. Okay. Thank you very much. Well, it's a great pleasure to be with you and with your colleagues at the European Central Bank and all the people that attend this seminar. I'm very honored to be with you and I'm going to present results of research with it. You may G and myself. By the way, you may not run the School of Economics, but University College London, which is very close to LSE as you may know. Okay. So I'm going to talk about trust and monetary policy and I call this a behavioral economic approach. So let me by way of introduction, tell you what I want to do. I want to analyze how trusts affect monetary policy and I will use a behavioral macroeconomic model. Its main characteristics are that we assume agents to have cognitive limitations that prevent them from having rational expectations. They only understand small bits and pieces of the whole model and as a result, you use simple rules to guide their behavior. But there are no rules and we introduce rationality to a selection mechanism and this selection mechanism allows these agents to evaluate the performance of the rule they are following and they will decide depending on their evaluation of the performance of these rules to switch to those rules that perform better. So the first part of my presentation will be to present very briefly the model and then focus on trust, how we define trust and what implications are for monetary policy and finally I'll say a few things about its relevance for today. So let me first talk about the basic structure of the model. So it's a very simple model in a way. It's a New Keynesian model. It has an aggregate demand equation that explains aggregate demand a forward-looking component that is the expected future output gap and also the real interest rate is derived from utility maximization. Then the second component is an aggregate supply a New Keynesian Phillips curve that explains the inflation rate and there again there's a forward-looking component. Agents make a forecast about inflation and that then affects the inflation rate observed today and the output gap appears there as an exploratory variable when there is a positive output gap that will tend to stimulate inflation and vice versa. And then finally the Taylor Rule describes the behavior of the central bank that sets an inflation target but also is concerned about stabilization of the business side of the output gap. So that's a very conventional basic structure of the model that we find in most microeconomic models these days at least the simplified versions of microeconomic models. What is less? Whoops, just a second. So the originality of this paper that is to introduce behavioral assumptions about how agents make their forecasts. So instead of assuming rational expectations the same agents can not easily do because of cognitive limitations they use heuristics, simple rules. And here we reduce the model to its bare simplicity that is we assume just two rules. One we call a fundamentalist rule where agents forecast the output gap to return to the steady state in the next period. We can have more complicated rules that we have experimented here with but here we assume the simplest possible rule. And the second rule is an extrapolative one. Agents, according to that rule, agents extrapolate the past output gap. Note the following in the fundamentalist rule you have a negative feedback rule agents will tend to forecast the output gap to go back to equilibrium, right? So it's also a stabilizing rule while the extrapolative rule is a positive feedback rule and it is the interaction between these two types of rules that is responsible for the complex dynamics that will come out of this model. Something similar is done with inflation forecasting. Again, two forecasting rules. One we also call a fundamentalist rule although it's a little bit different in nature but here we take the view that the fundamentalist rule is the rule whereby agents use the announced inflation target of the central bank as their forecasting rule. In other words, these are agents who trust the central bank. They believe the central bank and therefore if the central bank announces an inflation target of 2%, they believe that in the next period inflation will go back to 2%. The extrapolative rule is then the other one that is just agents that extrapolate the past. So in a way you can say that if a lot of agents use the extrapolative rule, they don't trust the central bank. So we will have here a way to measure the degree of credibility and trust that agents can have in the central bank. Now, these are the only equations that I can show you and all the other equations I eliminated but the market forecasts then are defined in the two equations that you see there. The market forecast of the output gap is a weighted average of these two rules that I just talked about, the fundamentalist rule where you see ETF, the fundamentalist rule and the extrapolative rule, ETE. And these are the weighted averages and the alphas are then the probabilities that agents will use either the fundamentalist rule or the extrapolative rule. And we have something similar for the inflation forecast, the market forecast is also a weighted average of these two rules where these beta-ardental probabilities that agents choose the fundamentalist or extrapolative rule. We can also interpret these alphas and betas as the fraction of agents that use these particular rules. And so these alphas and betas are determined by the performance of the rules and agents are selecting the rules that forecast best so they will switch from one to the other rule when they find out that the one that they are using at a certain moment does worse than an alternative. And so that's the mechanism that will drive these alphas. We use discrete choice theory to derive these alphas, but I'm just giving you the intuition here these alphas will reflect how agents adjust the rules depending on how well these rules have been doing. Let me quickly define a particular concept that plays an important role in this model. This is the concept of animal spirits. You remember what Cain said about animal spirits. These are, in fact, it's an index of market sentiments that we derive from these alphas that I just talked about. We can just transform these alphas into such an index of animal spirits, which we call ST here and that can vary between minus one and plus one, minus one will be a situation where agents, all agents expect a decline in the output gap, ST equal to one, all agents expect an increase in the output gap. In the first case, we will say pessimism prevails. All agents are pessimistic. When ST is equal to one, all agents are optimistic and it can vary between the extremes. For example, when it is zero, then there is neutral neutrality, optimists and pessimists tend to cancel each other out. So this is an important concept in our model to which I will return. We calibrate this model using numerical values of the parameters that we find in literature. We do a lot of robustness that I have no time to talk about here and we simulate it because it is a highly nonlinear model using IID shocks, so there is no structure at all in the shocks in aggregate demand, aggregate supply, and the Taylor rule, it's all IID normally distributed with sound deviation assumed to be 0.5, which calibrates the model to represent typically quarters. Given the size of the shocks that we apply, we can interpret these time periods as being quarters. Let me just say in one minute what this model produces as results and the key result of this model is that it produces endogenous business cycles. We don't have to rely on shocks to produce dynamics of business cycles. And the model predicts that a sequence of booms and busts that can occur in an unpredictable way and there can also be periods of tranquility where very little happens and all this is made possible by these dynamics of animal spirits that have a self-fulfilling property. Optimism can start developing and that leads to a self-fulfilling process of increasing economic activity that then leads to more optimism and leading to booms and busts. And as a result, also we find in this model that the distribution of the output gap is non-gaussian. It exhibits excess kurtosis but also fat tails that occur and lead to deviations from normal distribution. I will also have the opportunity to come back to this. So let me now turn to the major topic of this paper that is trust. How do we define trust here? And we have two dimensions to trust. One we call an institutional one. This is the trust in the central bank that has announced an inflation target. Do agents believe the central bank? Do they trust it? And actually as I told you before we have a measure of this trust in this beta FT. This is the fraction of agents that use the inflation target announced by the central bank as their forecasting rules. So these are agents who believe the central bank. They think the central bank is going to make sure that inflation goes back to 2% if that is the inflation target and the fraction of agents that use that as their forecasting rule is this beta F and we will use that also as our measure of trust in the central bank. The second dimension of trust is trust in the future and here we will use this index of animal spirits. The ST that I mentioned to you that measures the degree of optimism and pessimism about the future economic activity. We can get into a situation where everybody has become optimistic and see the future to rosy but this can turn around into pessimism and this index ST that will measure this dimension of trust. So the way we analyze the importance of trust is in the following way we apply shocks in particular we will focus mostly on supply shocks as this seems to be a shock that has been important recently and we also focus on big shocks by that I mean shocks that are really very large I will come back to that but I mean practically this will be contrasted to very briefly with demand shocks and I will also do some sensitivity analysis about the importance of the size of the shocks. So here it is I will present to you impulse responses to the supply shocks now this is nonlinear model and therefore the impulse responses very much depend on the initial conditions the exact moment the shock is introduced and in order to illustrate this we will compute 1000 impulse responses to a large supply shock that occurs at a particular point in time but with each new impulse response we compute we use a different realization of the stochastic shocks in the model remember the stochastic shocks in the model are the stochastic shocks in the demand equation the supply equation and tail rule so we run 1000 times different realizations of these stochastic shocks and each time at the same point in time we apply the shock which is a very large one which is a 10-style deviation shock now you may say that's indeed very large yes it is but this corresponds to actually what we had in 2020 with the pandemic it comes close also to what we had during the financial crisis of 2008 and as Carmen earlier indicated these shocks were indeed historically very large also these impulse responses are expressed as multiplies so let me now show you one of the major results here this is impulse response to a large negative supply shock 10-style deviation so this supply shock occurs in period 100 suddenly it occurs in period 100 and what you see the impulse responses there was a great variety in these impulse responses but the major result that I want you to focus on is that there seem to be two trajectories one we have colored green that's a good trajectory and the black one is a bad trajectory and you can see that in the bad trajectory the response of output is much more negative and also it takes longer for output to go back to equilibrium the same or something similar is true for inflation you can see that in the bad trajectory the shock is higher and it takes longer for inflation to go back to equilibrium as compared to the green, the good trajectory and finally something similar happens with the interest rate note that in the bad trajectory the central bank is actually increasing the interest rate in a much stronger way than in the good trajectory so these are the main features here surprising bifurcation the same shock occurs at the same time but initial conditions are different and leading to two very different trajectories but not also within each of these trajectories there's a lot of uncertainty, there's a lot of variation it's not that you can predict perfectly but surely there is this distinguishing feature of a bifurcation that can also be seen in the following way in this figure I show you the impulse responses 12 periods after the shock so I've actually taken a cross-section of the previous graph in period 12 after the shock and then you obtain these impulse responses and you can see the bifurcation there are two peaks there so when you look at the histogram of the short-term output response you have a concentration about around minus 1 minus 1.2 and then a concentration much more benign around minus 0.2 so you have this bifurcation that I talked about and something similar occurs with inflation in the back trajectory you have a lot of inflation and in the good trajectory inflation actually is around 0 after 12 periods after the shock and finally something similar is to be seen with the interest rate so that's another way to look at it I'll tell you how it allows you to see what the nature of the bifurcation is so the question that arises why do we get these bifurcations what's underlying this and here is the back trajectory is characterized by the fact that immediately after the shock we obtain a limit solution that is inflation credibility drops to 0 and atmosphere drops to minus 1 and that means that the mean reverting processes the negative feedback rule in the expectations formations are switched off and only the extrapolating dynamics the positive feedback rule is left over so that's the fundamental reason so you have a very large shock that brings you to the limit solution extreme pessimism total loss of trust in the central bank and as a result the mean reverting forecasting rules have disappeared and everybody is extrapolating and this introduces a destabilizing dynamics that keeps the output gap low and inflation high now how are these trajectories connected to our measures of trust I have defined trust earlier and so what I'm going to show you now is the evolution of my two measures of credibility one is our own spirits and the other one is the fraction of agents that believe the central bank and since we run this model 1,000 times we obtain 1,000 trajectories of these two variables so here they are let's first concentrate on the black these are the measures of trust in the bad trajectories and look at the one on top here inflation credibility with shock bad trajectory note that so up to period 100 nothing is happening and you have this all black there because 1,000 little lines that move up and down but then in period 100 you have this shock and you can see that immediately or very soon after the shock credibility inflation credibility drops to zero nobody believes the central bank anymore and it takes a while more like 12 periods this is about three years for credibility to reemerge where this becomes positive again these fractions become positive again and something similar occurs with annual spirits immediately after the shock everybody becomes pessimistic and it takes a long period for pessimism to disappear and optimism to emerge in the bad trajectory in the good trajectory the green ones you don't see much with some movement of animal spirits and also of credibility but there is not much I mean in terms of loss of credibility most of the time credibility both in the central bank and some optimism is maintained so that's the big difference between the two trajectories in the bad trajectories trust is lost and that's the characteristic feature of a bad trajectory loss of trust in the central bank loss of trust in the future of the economy what is the role played by initial conditions in this model well the initial conditions are key so actually this is the question of why do we get into bad trajectories in the first place how does the economy get into a bad trajectory and here the answer is well initial conditions must be bad what are bad initial conditions first high inflation expectations and second low inflation sorry output expectations so high inflation expectations means that agents expect inflation to be above the target of the central bank and these bad initial conditions then make it possible for the large negative shock to push the system towards the limits of zero credibility and extreme pessimism it is because initially you were already in a bad state that this large shock quickly brings you to the limit and then reinforces this and leads you into a bad trajectory when on the other hand the initial conditions are favorable these inflationary expectations are low there is some optimism about the future then the same negative shock will not bring you to the limits of the system and as a result mean reverting processes continue to do their work of softening the impact of the supply shock and one that ends up in a good trajectory so in other words these initial conditions favorable initial conditions work as a buffer preventing large shocks from hitting the boundaries and preventing a collapse of trust because it is initially because things look good right then the negative supply shock well doesn't push you all the way to the negative limits that the system then can end up in a good trajectory and then you can see that trust here is key in smoothly returning the economy to equilibrium if in a good trajectory trust is maintained and therefore you go back to equilibrium faster than in the bad trajectory so that is the key message that these models give us let me quickly say a few things about negative demand shocks I'm not going to go into detail I'm not going to show you actually any impulse responses we do that in our paper or by the way before I go to the negative demand shock let me say that in our paper we also do much more precise analysis of how initial conditions can forecast the future trajectories we do some econometrics and all that to analyze this and we do find that these initial conditions especially the expected inflation is a very good predictor of whether or not you will go into a good or bad trajectory in this model so negative demand shocks then also large negative demand shock we get a similar bimodal distribution but much weaker so there is not much difference between good and bad trajectory what is the explanation well here this in contrast to a supply shock a demand shock does not put the central bank in a dilemma situation I don't have to tell you what that means a dilemma situation right in a supply shock the central bank is in a dilemma situation because basically the interest rate will lower inflation at the cost of recession and vice versa if you want to prevent a recession when you have a demand shock you don't have this dilemma situation and as a result what the central bank is doing is that it can do something effective by lowering the interest rate for example after a negative demand shock it works both in in raising inflation again and raising output again and as a result the central bank is seen as an institution that performs well and trust is maintained so there is very little decline in trust after a negative demand shock compared to a negative supply shock where the major problem is that trust can disappear okay quickly about sensitivity analysis the size of the supply shocks what I've shown you the result is when a supply shock has a standard deviation of 10 so what we do is to reduce the size of the supply shock so you can look at this and move upwards and you can see that as the supply shock becomes smaller this bifurcation tends to disappear there is still a lot of variation and when you look at the second column of the impulse responses the frequency distribution can see that there is no normal distribution certainly not when the supply shock is very big but even when the supply shock is relatively small you don't get anything near a normal distribution which leads to problems how do you about how to forecast what these impulse responses to great uncertainty about all this the power of output stabilization let me say a few things about output stabilization when you have a supply shock the results that I've shown you we set the output parameter in the Taylor rule equal to 0.5 which we call the normal case the Taylor rule with an output parameter which we call C2 equal to 0.5 seems to be also what central banks, at least in econometric results that we have seen seem to have do now is to contrast impulse responses that we have shown you with a normal C2 with one where we assume much stronger output stabilization we set C2 equal to 2 so it's a much stronger output stabilization what does that do to these impulse responses after the same supply shock and here is the contrast on the right hand side these are the results that I showed you earlier which what we call normal output stabilization C2 equal to 0.5 when stabilization is very strong the central bank is very ambitious in trying to stabilize output after the negative supply shock we can see that yes output stabilization is very successful it reduces the downward movement and also it tends to reduce the difference between the good and the bad there's still a good and a bad trajectory but the difference between these two has declined significantly but there is a price to be paid for this success not surprisingly here I show you the impulse responses of inflation on the right hand side again these are the results that I showed you earlier and on the left hand side I show you impulse responses when stabilization is very strong when C2 is equal to 2 and what you can see the contrast here is that inflation gets ingrained it takes a very long time for inflation to go down right it's like it becomes endemic in the system both in the case of the good and bad trajectory of course it's even worse in the bad trajectory but in the good trajectory you can see that there are many green lines that take a long time to go back to a killer there are some that do it fast also right you can still have if you are lucky if your initial conditions were good enough you can still end up in a relatively benign trajectory but probability that this happens is relatively small you can still find here and then finally what happens with trust under these two output stabilization rules on the right hand side I show you the inflation credibility and animal spirits under normal stabilization these are the results that I already showed you earlier and now contrast this with what you obtain under strong stabilization and the striking feature is that the central bank is much longer look at this graph inflation credibility in the bad trajectory when you have strong stabilization the central bank is very ambitious in stabilizing output well the loss is also immediate but remains very low for a very long time so trust is lost for a very long time something similar occurred with animal spirits conclusion so we conclude that negative supply shocks create important threats to trust in the central bank and in the economy all the more so when central banks pursue aggressive policies of output stabilization this is much less the case with demand shocks for the reasons that I have given you right but the fact that when there is a negative demand shock there is no dilemma and therefore the central bank is perceived as being successful with a supply shock central banks are in a dilemma situation that prevents them from successfully destabilizing the economy and if you then try harder it only makes matters worse so that's a key insight some relevance of our results let me conclude with that does all this have some relevance here it is during the 1970s we had large supply shocks so I will contrast the 1970s with the more recent Covid supply shock and when you look at the 1970s there had already been a significant build up of inflation there were significant expectations when the supply shock occurred in fact this happened during the second half of the 1960s and then our model predicts that given these unfavorable initial conditions the recovery would take a long time and this is exactly what happened in many countries that had a prior history of significant inflation especially after the second oil shock of 79 it took a long time for the world economy to recover this was not the case with the supply shock of 2020 which was preceded by a period of very low inflation and low inflationary expectations and then our model predicts that the recovery could be quick and that's exactly what happened it also allowed central banks that did not have to worry much about inflationary expectations to actually follow expansionary policies in 2021 was quite strong but then unfortunately a new shock occurred in 2022 and here what does our model tell us that's a final not very uplifting thought for you when the Ukraine shock occurred the initial conditions have deteriorated significantly and this creates the risk that we may be hitting a bad trajectory in the future with a strong decline in trust high and long lasting inflation a deep recession and central banks that are forced to increase interest rates to very high levels thank you for your attention thank you very much Paul for this very nice presentation a bit less uplifting message at the end but this is the reality that we are facing Natasha Vala she's currently at CSPO but she's a very well known phase at ECB she was of course working in the multi-policy department also at ECB not so long ago thank you very much Isabel and thanks for the invitation to this conference which is close to my heart I have to say and I guess you know that I'm sharing my slides now to comment the paper and thanks Paul for the presentation I have to say that after your talk I see the paper a bit differently but I still find it paradoxically quite refreshing despite the gloominess of the charts that you showed us and the gloominess of your conclusions and refreshing why because it's another way literature that we have in monetary economics in generating and the genius business cycles and also that revisits in a sense at least that's the way I see it this view of sunspot equilibria and multiplicity of steady states and being careful about not to fall in the wrong one and why initial conditions matter and how a novel approach and I think it's a good start for a new trend in the literature so I start with a very short summary of what I think the originality is in this paper and then a couple of questions to which you may want to reply if we have time or we can also let it to the general discussion so trust as you said is here at the centre it's a factor in the transmission of you comment mostly negative supply and demand shocks there are two levels of trust one is regarding the policymaker itself and the second one regards the strength of the economy and interesting thing I'll come back to that is that you have a bimodality in the outcomes that good and bad outcomes are possible as a response to these shocks in particular at the extremes there's a sort of endogeneity of trust because those betas are the result of interactions but there is an asymmetry in the relation between the outcomes and the dynamics of trust so this asymmetry this is where I have a couple of questions that are also related to this what I tend to see being Bayesian I see that it's a moving state which then for some miracle comes out of it again but that's another way to think about it it's also a framework where initial conditions matter I said it already so there's no you know unicity you have past dependency in some sense you don't need to worry because of the instability of the equilibrium steady states that we the worry that we have usually in the frameworks where we have those things at the center of the equilibria mechanism this is a calibrated model you said it but that's fine because your model is helpful to derive stylized interpretations of outcomes that we see in real life so my first question but it's really a question the Bayesian I would say in French because I'm not a behavioral economist it is more on how to compare behavioral models applied to monetary policy how to compare them with learning models because it feels very much as your pool of extrapolative people are more towards adaptive you know and adaptive learning would be close to you know the cognitive limitations you highlight in your description and your fundamentalists would be more the rational learners at least that's the way I would see them and that's interesting to draw the parallel because the one and the other you know families of models learning models have different implications for policy actions some tell you you need to be cautious you need to underreact two shocks the brainer uncertainty affects those outcomes and others tell you you should experiment and you should be bolder in your reaction so I wonder if having a behavioral framework sort of doesn't go either one step further and tells you it depends on how the blend of the population is so I'd be happy to listen to you in connecting those two literatures now as you have specified as the model is specified in particular with the tailor rule it cannot really account for strategic interactions because the policymaker is not optimizing would there be a way without complexifying the framework to account for them because I feel it's quite important I'll come back to that in particular in current circumstances where credibility and the dynamics of credibility and that's at the core of your paper perhaps very much depends on the initial conditions but the initial conditions today are unique in history at least in the history of central banking now a second point I have a set of questions on the bimodal distribution of the impulse response functions you document a non-gaussian output gap you said it and there's a bimodality of the responses of output gaps and inflation and also interest rates actually so I was wondering and it really is a genuine question because when we have models estimating the output gap there's not so much in the literature that looks for these non-gaussian properties or at least that doesn't relate it to bimodal responses of the main variables and I think it might be worth it and the fact that you find those and they seem to be very robust in your framework shouldn't we go back to those models I remember when I was at DCB at the very beginning Isabelle you might remember that the work of Thomas Vesterman was working on it as well and at some point they were documenting those bimodalities and then we left that behind and we go back to it open question and here should probably be another slide really but you try to be optimistic in your conclusions now in presenting but I would be utterly pessimistic when I read your paper and when I compare your outcomes with the current circumstances because what you show and this is the draw line where you sort of drop and it takes a long time, 12 periods to perhaps come out of the state I mean I call it a state, it's not a state in your model out of the bad outcome again so when you have those bifurcations you fall into trap where credibility is down to zero if there's a really bad shock but what if, and that's right what you said at the very end of your talk you have a sequence of really bad shocks which is like what we are having right now and the sequence of very bad shock is kind of compounding in the lower output gap and high inflation configuration then it sounds very difficult to believe that with time memory goes off and credibility comes back again on the side of the policy makers either strategy or parameters in the reaction function I understand this does not feature in your model because you have for good reasons this rule but it sort of begs for this discussion at this point so given the asymmetry in the responses aren't we bound to be stuck in the bad outcome in the role of very negative shocks as we had and what is the policy implication then shouldn't we then overreact and then I'm back to the sort of learning prism of the learning to read the situation shouldn't we think about changes to the parameters of the reaction function that would be useful to have a discussion on this on yeah that's it's saying more or less the same thing Isabelle you tell me how much time I still have I only have a couple of slides left but if you give me another 2-3 minutes I'll be happy so there's one thing that is not really formalized you equate trust with credibility but again there is this ambiguity trust with credibility and trust with optimism and confidence and I think here it is key not to blur at least on the side of the trust that has to do with the policymaker ability to reach its objective not to mix this credibility and optimism interpretation of trust I don't know if I'm clear on that but I think in other modeling approaches that would be in the equation here it is not and those are two different objects and the last thing really the relationship with the optimal policy literature that's also a very tempting link that we would like to make and it would be useful for you to link connect your results with this in the current circumstances we also have but that's probably too far from what you're after the more specific question among supply side shocks of oil driven supply shocks because it would specify the supply shock as a relative price shock and that could be also interesting but I think placing this because this is also a very long tradition this optimal policy literature in relation to the shocks we see right now and inflation environment that we have right now so it would be useful to place it in this sort of in this tradition there's one paper which I found relatively relevant to your the question you're asking your paper it's the paper in the IJCB it's a bit old now but I think it's still valid of Berriesville and the Camille Cornon on transparency and communication what the optimal response to cost per shocks is and the fact that it may depend on the central bank communication and on disclosure in their paper both signs are possible so you can really go both ways and I think this link this interpretation of trust so on the side of the policy maker trust trust towards the policy maker might also have to do with communication with openness and with disclosure of preferences initial conditions matter as every economist say that's a very important point because you have this path this starting point dependency and probably also path dependency and it would be useful to have the economic rational you have it and in your presentation you were very clear so maybe I need to look again at the paper but that's also important it reminded me of the paper by Caballero recently that looks at temporary supply shocks so this is not something you look at the persistence of shocks but it's a good comparison point on those initial conditions because they look at it again as well and I'm done Isabel thanks for your patience and I will stop sharing this the screen thank you Thank you very much Natasha there was quite a number of questions and we have limited time but maybe I give the chance to Paul to react to some of the questions and I see if in the chat questions come up from the audience but please go ahead Thank you Natasha for your thoughtful comments I interpret many of these to be suggestions for further research and I think you point out the important issues that we want to discuss and analyze and research in the future your first point you ask is the question how to compare these learning models adaptive learning our model is a model of adaptive learning where agents try something right with rational learning I interpret that to mean statistical learning where agents use statistical methods to learn so it's interesting to make a comparison but one of the things that will make this comparison difficult is that we create a dynamic where we systematically deviate from normality and that makes it very difficult to implement statistical learning techniques because they are almost always based on normality assumptions another way to put this when you look at these impulse responses that we generate there the bimodality this creates a fundamental problem of ambiguity you have a shock but agents don't know what to do because the shock has ambiguity effect, ambiguous effects where shall we end up and that makes it difficult to apply statistical learning techniques in my view and maybe the more rational approach is to have adaptive learning where you say I don't understand the world sufficiently and I observe that the world is not normally distributed maybe we have to rely on other learning methods but here I don't want to be dramatic and if you can convince me that we should use these other learning methods that's ok with me you asked the question what about the sequence of bad shocks actually we looked at that recently we have now done simulations we are preparing a new paper where we have a covid Ukraine shock we call it and yeah that's bad news I can tell you it's not funny look at these shocks as you mentioned you are quite pessimistic what might happen in a model when you have a sequence of bad shocks in that way I take your point about trust in a way we are a little bit loose about the definition of trust we have two dimensions there one we call credibility, optimism we have been talking for a while about this but we separate we always separate credibility from the animal spirits and they do not always go in the same direction that was not clear so much in the supply shock results that I showed you but when we introduce demand shocks large demand shocks central banks have very few problems with losing credibility they maintain that credibility throughout and the problem arises more than from the animal spirits the market sentiment if you have a very negative demand shock that can lead or can create lots of pessimism that is self-fulfilling that puts you into a bad trajectory while the central bank continues to maintain its credibility in terms of inflation so we do make that distinction it's not that we all put it together and it's mixed we always take care of separating that your point about optimal policy literature and a connection with that also I guess for the reason that I already discussed optimal policy literature is very much formulated in the context of a world where shocks are normally distributed that makes it easy to do optimal policy analysis once you depart from that it's very difficult to design methods to do that because the essential ambiguity you have a shock and you don't know what the nature of that shock will be is that bad news or good news and actually that's what we have seen with central banks when the shock occurred any central banks didn't know what to do they should be worried about this in fact there are two different scenarios possible or maybe even more so and that's what our model tells us we have a large shock and there are different scenarios possible but in real time when it happens you don't know which scenario will actually come out of all this and that creates ambiguity making analysis of optimal policies very difficult but Natasha thank you for your comments Paul meanwhile there were two questions also in the Q&A chat so I will just read them out for you one comes from Mathias Farkas from the ECB starts by thanking you for the excellent presentation and he mentions you introduced two non-linearities animal spirits and trust in the central bank and he asks should we not see four regimes from the bio-modality it seems that the regimes are related does it mean that central bank credibility is susceptible to animal spirits and vice versa so that's the first question and the second question comes from Abhishek Das he also texts you for the lovely talk and he says one of the crucial aspects of the adaptive learning that you were mentioning is understanding the network components or rather the interactions between the network components however when we understand the new shocks arise these interactions are not necessarily quantified or largely identified very accurately on the system due to underlying factors how should we approach using such models for policy decisions which could be harsher or less harsher depending on what our models tell us so these are the two questions and I had a question myself I mean we are with squeeze on time but followed a great presentation and I was wondering also I mean trust is an unobserved variable in a way right but you know we have surveys on trust and have you tried looking at you know following certain shocks whether we see in survey responses moves in trust that would actually corroborate what you find in your model and another question I mean I guess it's maybe more for future work is I mean you know sometimes we say once trust is lost it's lost forever and you don't really find it at your model but as Natasha mentioned it takes a long time to regain it but you know what kind of policy maker do to regain the trust and I think Natasha had something similar like you know should the responses be for instance more rapid more stronger to certain developments basically to regain trust or what should the policy maker actually do or what can he do? Okay thank you very much yeah on the interaction that's a good question interaction of the two dimensions of trust looked at it in a very independent way right you have trust in the central bank and animal spirits market sentiments optimism and pessimism and they have a kind of a life of their own of course there is an interaction in the dynamics of this as you can see from the evolution of these variables after the shock they do interact with each other but we haven't really analysed systematically the nature of that interaction that's a good suggestion to look into that on the second question network components of course this is a very aggregative model so I'm not sure I can say much about this because networks you would actually have to go to a granular model right with where you have interactions between many agents and that have a network quality I think in agent based models that is more appropriate setup to analyse these questions that I don't think our model is capable of doing well thanks for your comments we have looked into surveys and they exist but we have not done any systematic work and we do plan to do that and also in your last question what can policy makers do to regain trust when they have lost it so this is also a question that we are very much interested in maybe I might suggest that the ACB might be willing to sponsor such research thank you for your attention thank you very much Paul for the great presentation for Natasha for the excellent discussion it was great to see this and it's definitely a food for thought for future work I mean trust is very important especially central banks so you're always interested to learn more and understand how this can change due to shocks and what can be done to mitigate it