 the speakers for today. So Matthias is a professor of macroeconomics at Goethe University and a research fellow at the Halle Institute of Economic Research. Matthias Discussant will be Anuka Ristiniemi, a colleague of mine at the ECB, who is team lead economist in the Monetary Policy Strategy Division. And Anuka has prior to her start at the ECB worked at Rix Bank and Bank de France. Unfortunately, Moritz Schulerich, who will give the second paper, is late today. So, but he, I was told that he will arrive in a minute. So he is president of the Kiel Institute for the World Economy and professor of economics at Sciences Po and research professor at New York University. He also received the Leibniz and the Gossen Prize to very prestigious German research prizes. And the Discussant will be José Luis Pedro, who is professor at Imperial College London and professor also in Barcelona and has close links with the central banking community and has different roles at the Bank of Spain, the ECB and the ESRB. So, so much for the introduction. Let's start with the first paper by Matthias, which is on understanding post-COVID inflation dynamics with Martin Harding and Jesper Linde and the floor is yours. So, 35 minutes. Thank you very much. All right. Let's see. Could I ask you to put up my slides, please? Thank you. Okay, go back one slide. This is not the right beginning. Could you scroll back one slide, please? Thank you very much. All right, excellent. Good morning, everyone. Thanks a lot to the organizers for having me. It's a great honor and pleasure to be here today. The work that I'm presenting is joint work with Jesper Linde from the IMF, who is here with us today too, as well as Martin Harding from the Bank of Canada. The usual disclaimer of course applies. Now, what are we grappling with in this paper? Well, this is a time-series plot for core inflation measures in the United States, in the Eur area, as well as in Canada. And if you look at that time-series plot for the sample plotter, there are at least two major puzzles that have caught the economist off guard. The first one, indicated by the first vertical line, is the financial crisis and ensuing great recession, where many economists have expected, or had expected at the time, deflation to happen and to occur as a fallout of the gigantic crisis that did take place. Now, when you look into the inflation figures presented in this graph, there was no deflation, and it remains a puzzle to many economists why we didn't see a deep fall in inflation in those periods. The buzzword is the missing deflation puzzle here. The second big puzzle that has caught macroeconomists and policymakers alike off guard is indicated by the second vertical line, which is the pandemic. And after the pandemic has waned, inflation across many industrialized countries has basically taken off like a rocket. And many policymakers, academics alike, were puzzled how quickly inflation actually took off. What we aspire to do in this paper is to study those two episodes and to offer a unified framework, a way to think about these two events in a unified model. So we don't aspire to write a different model for the financial crisis and one for the post-COVID. Now, we wanna work with a unified model that has the ability to come to terms with accounting for the two key puzzles in terms of the missing deflation, as well as the rocket type takeover of inflation. All right, so how do we do this? And what sort of say the game that we are playing here? Somehow I have difficulties advancing my slides. If I just say next slide, please, is there a way that you could forward next slide, please? Thank you. Something is, here we go. All right, so the challenge just to be really clear is to reconcile the missing deflation puzzle of the financial crisis with the recent surge in inflation. How do we do that? Well, we focus on the US in this paper and we study inflation and output dynamics using the workhorse model that Frank Schmetz and Ralph Vauders have put together. Now, we're gonna put a spotlight on one particular feature of that model. And that feature is a Kimball state dependent demand elasticity. It's a feature that's embedded in the Schmetz and Vauders model. It hasn't been utilized as much as we like it to be utilized. And so this is what this paper is essentially all about. After everything is said and done in our paper, what you're gonna see is that the Phillips curve becomes state dependent in the sense that the Phillips curve is not a linear relationship, but it's a boomerang or banana type shape Phillips curves that we know that the model then entails. And we're gonna look into, you know, how do the shocks that hit the economy propagate in such an environment? All right. Now, for those who maybe still a little bit tired because of the early morning, let me just give you a preview of the results that I'm gonna present in the next 30 minutes. We show that our variant of the original Schmetz and Vauders model explains the modest decline in inflation during the Great Recession, as well as the post COVID trajectory of inflation much better than the original, and I should add here Schmetz and Vauders or linearized Schmetz and Vauders model. What is particularly important is that the non-linearity embedded in the Kimbell aggregator is something that we put at full force when working with the model. So the non-linear formulation of the model is really important to arrive at our key results. So put differently, had we worked with the non-linear Schmetz-Vauders model during the financial crisis and during the post COVID in a rocket trajectory, this model would have given us a much easier time coming to terms with the inflation dynamics that we had observed, that we have observed. Now, at the core of our analysis is that the Phillips curve becomes steeper during booms and it becomes flatter during recessions. So that's gonna be a feature that the model entails. Now, importantly, once you think about how shocks propagate in such an environment, we document that cost push shocks or exogenous variations to firms' marginal cost and you pick your favorite paradigm. You can think about these are changes in energy prices that the model models in an exogenous reduced form way or this is China zero COVID type supply side change disruptions. These cost push shocks that hit the economy, they are propagating in the economy in a different way depending on the state of the economy. In a boom or when inflation is high, a given cost push shocks adds an additional amount of inflation that's much larger than the same size cost push shock when it hits in a recession. So the effect of cost push shocks on the rate of inflation are inherently state depending in the framework that we are studying. Now, with this all in mind, this has of course important implications for policy making in the following way. The policy trade off between stabilizing inflation and output in response to cost push shocks becomes steeper if inflation is elevated to begin with. So if inflation has left the bay, if inflation is above target and substantially above target, it becomes increasingly more difficult for policy to stabilize inflation and output as compared to when inflation is cruising around the long run target that the central bank may aspire to hit inflation. All right, now let me put a little more flesh on each of these four bullets and walking you through the paper and the analysis. We employing the workhorse macroeconomic model that is used across many central banks and policy making institutions, including academia in particular, we're using the Smets-Vaudras model whose ancestor is the Cristiano Eichenbaum and Evans model. Now, importantly, we work with the non-linear formulation of that model. I'm not gonna drag you through the details of the model. I take it there are many ECB economists here that have at least heard about the model if not have worked with the model, okay? Where we are gonna put some spotlight on is that we're working with the Kimbell aggregator here that is embedded in the Smets-Vaudras model. Now, if you happen to have forgotten about what that Kimbell aggregator does, let me just give you in a way that Kimbell aggregator is sitting in terms of the model. There are competitive firms and I like to think about these firms as, say, restaurants putting together restaurant meals and these competitive firms, they aggregate intermediate goods YF, okay? So these intermediate goods YF are lettuce, tomato, bread, wine, beer, you name it, and they have a production technology to aggregate these goods into a restaurant meal. That technology is a Kimbell technology. That's a Kimbell aggregator, okay? Now, we're gonna study the implications of working with this Kimbell aggregator. Now, since we wanna work with an explicit non-linear model, we need an explicit, you know, functional form for that model. And we're working here with the specification by Dotsy and King in the QGE, as well as Levin, Lopez, Salido and Yun. I'm not gonna drag you through this rather complicated looking formula for the Kimbell aggregator. But there is one key parameter that allows us to actually bridge the gap to something that you may are more used to working with. The parameter psi is of particular importance here. If you set psi equal to zero, you're back in the Dixit-Stiglitz type aggregator word. Well, that many of you may have heard and you are familiar with from, say, textbook monetary economics texts. If psi is non-zero, you're gonna enter the Kimbell world. Now, what's so special about the Kimbell world? Oh, my slides have disappeared, unfortunately. It's kind of important to have the next slide on. Thank you very much, appreciate it. All right, now, so what's so important about this, you know, Kimbell aggregator, let me give you the intuition about the Kimbell aggregator in one slide. And if you understand this slide, you have understood 95% of the paper, so it's downhill after this slide, all right? So with Kimbell aggregation, what you're gonna get is that the price-setting problem of monopoly price-setters becomes asymmetric. In particular, it has a feature that once you work with Kimbell, the price-setting problem becomes quasi-kinked in terms of the implications of the demand curve embedded in the model. Now, let me, so to say, offer you a static example of Kimbell aggregator, you know, optimal monetary policy price-settings. So take the Smets-Vauders model and strip away, you know, price rigidity, wage rigidity, capital. So just take off all the important features from the Smets-Vauders model. What you arrive at is this static monopoly power price-setting problem. That's at the core of the model. A period-by-period price-setting problem for a monopolist. When you work with the Kimbell aggregator, you're gonna see that the Kimbell aggregator gives rise to strategic complementarities in the following sense that the demand elasticity for these intermediate goods, tomatoes, lettuce, beef, beer, wine, they are an increasing function of a firm's price. So if a firm increases its price, the demand elasticity increases. So that is, more and more customers are gonna leave the store. That's basically the idea, okay? So a firm increasing its price, it's gonna lose proportionally more customers as it keeps on increasing, compared to the Dixit Stiglitz case. Now, even with this in mind, it turns out that firms increase prices sharply when marginal costs increase. How is that possible? Well, if marginal costs increase, so think about an energy price shock, think about zero COVID in China, your favorite examples, that squeezes profits of firms and firms have an incentive to raise their prices. Now, they rise prices by more than an Dixit Stiglitz for the following reason. Take a look at the left subplot here. That's a stylized way of plotting the demand curve for intermediate goods. It has the feature that the demand elasticity is state-dependent and increases with the relative price of a firm, okay? If you increase the price, you're gonna lose a lot of demand. What this implies is that this has implications for marginal revenue of a firm. So a monopolist seeks to maximize profits and it does so by setting marginal revenue equal to marginal cost. Now, the middle plot plots the marginal revenue curve implied by the optimal profit-maximizing problem of the firm subject to the king-demand curve that we see on the left-hand side. The blue line is the marginal revenue curve and the key implication or the key feature of that curve is it's concave, okay? It's concave in the firm's price. As a firm increases the price, it's losing more and more customers, so revenue at the margin is gonna get smaller and smaller at the margin for increasing the price, all right? Now, with Dixit-Stiglitz, the blue line would be simply a linear in-lock in our specification. If you, for a second now, think about what is the profit-maximizing price of this firm? Well, it equates marginal revenue. The blue curve weighs marginal costs and for a moment just consider a level of marginal cost depicted by the black dashed dotted line, which is horizontal here, the intersection of marginal revenue, marginal cost, that gives you the optimal price that the monopolist would choose. Now, consider variations in marginal cost. Consider a 10% rise in marginal cost depicted by the red broken line. What you're gonna see is that the price goes up, so the firm charges a higher price and charges a substantially higher price. Let's consider the exact opposite, the same size change in marginal cost, namely a cut in marginal cost depicted by the pinkish dotted line. You see there firms are gonna cut their price, but now if you compare on the horizontal axis, on the price axis, you compare the absolute changes of prices, you see that firms increase their prices more than they decrease their prices. That's the key feature that the Kimberleiger Gator brings home for us. And we're gonna utilize that from now on in the Smets and Bowders fully-fledged model. So in other words, if you do this on a continuous change of marginal cost, the far-right plot depicts that the optimal price is a convex function of firms' marginal cost. So if marginal cost go up, firms are gonna increase their prices more than if marginal cost go down. This is exactly what we're gonna lever on in terms of our analysis, in terms of addressing the missing deflation and the COVID rocket-shaped takeoff of inflation. All right, if you made it through here, you made it through 95% of the paper. From now on, it's just applying that in a fully-fledged Smets and Bowders environment. So let's put back price rigidity, let's put back nominal rigidity in wage setting, let's put back capital, let's put back habit-persistent, et cetera. So the standard elements that we work with in industrial-sized New Canes in models. When you then linearize the model, you're gonna get an object that many of you have seen in grad school, which is the New Canes in Phillips curve, which on the left-hand side has the quasi-difference of inflation as a function of the right-hand side quasi-difference and inflation expectations, marginal cost, as well as an exogenous disturbance, what we typically call a cost push shock. Now, where does Mr. Kimball come in here? Well, Miles Kimball comes in here via the slope coefficient kappa. So kappa is the slope of marginal cost and the Kimball parameter or the Kimball aggregation parameter comes in here in the basement or in the denominator of that kappa. So with Kimball aggregation, you typically get a smaller kappa. That's what we learned from Raph, Bowders and Frank Smets in their 2007 AER paper that with Kimball aggregation firms changed their prices by less and it helps the model come to terms with data. That's beautiful. Now, what we do is we're not working with this linear formula, rather we're working with the non-linear formulation of the model, okay? So now how does the Phillips curve look like in a non-linear environment? Well, it's a set of not so pretty-looking equations in fact. It's not much to write home about, except for that you can represent the optimal pricing problem in a set of non-linear recursive equations. So these equations are recursive so you can type them in in your favorite non-linear solver package quite easily, okay? If anything, marginal costs are showing up now in the third equation in a multiplicative way with the markup shock. So the shock that we're gonna highlight in our analysis going forward shows up in a multiplicative way. Of course, it works through all the other three non-linear equations in generic equilibrium. By contrast, if you look at the linearized Phillips curve, you see that the markup shocks is additively separable. That's gonna play a role going forward in our analysis. All right, so how are we gonna work then with this model? I'm gonna spare you the details on all the other elements of the model. They're completely standard. And we're gonna work with this model by first parameterizing it and then solving and filtering data with it. Now we're gonna follow our earlier work in 2022 in terms of parameterizing the model. We're estimating the model on, and that's important, pre-financial crisis data. So we don't allow the model to neither see the great recession data nor see in terms of parameters the rocket takeoff inflation data post COVID. The model doesn't see any of that data. So the parameters are informed by using 15-year-old data, outdated data. So we really put the model to a hard test. Is the model able to come to terms with something that the model really hasn't seen before? There's a little bit of solutions. I'm wrangling on the parameters. Long story short, we are estimating the Kimbell parameter in the model. So there's a way to estimate that and the paper provides you with all the details. We then use that model and we are solving that model and we filter data with that model. How are we solving that model? Well, we solve it with the so-called extended path method that's embedded, for instance, in Dynier. Goes back to Fair and Taylor, 1983, Econometrica. Some people may have heard about the time-stacking algorithm or the two-point boundary value algorithm. These are all the same things that we utilize and it comes very handy in terms of Dynier. Now we have all our codes available online on my website, for instance. So if you wanna work with this non-linear model, it's really grab and go and hit a button. We then could do stochastic simulations with the model and the only thing that we impose here is in certainty equivalence as a element that the solver imposes here on us. Now importantly, we're gonna show you now the differences between the linearized model and the non-linear model implications. So I'm gonna take you on a tour now what are the implications of working with the non-linear versus the linear model. Let's get going first on how do Phillips curves look like in this environment? And what you're gonna see is that with quasi-king demand, the Phillips curve become convex in terms of the output inflation space. So let's consider for a moment, you're hitting the model only with the man shocks, the left-hand plot, and on the x-axis you have the output gap and on the y-axis you have inflation. You only hit the model with the man shocks. So shocks that move output and inflation gap in the same direction. What you see is in the linearized model, you're gonna get a linear relationship. That shouldn't be a big surprise between the output gap and inflation. It's depicted by the blue dots in this scatter plot. Now, if you take the non-linear formulation, you're gonna get what we call a banana-type Phillips curve or a boomerang-shaped Phillips curve. You're gonna see there's some curvature in there. Now that curvature is gonna have very important implications for addressing the challenge that I put up here on my first slide. Particularly, as you go in recessions, so as inflation, when in the output gap goes toward minus five, minus 10, as you go into the recessionary area, you're gonna see that the red line starts to curve and bend and inflation does fall, but it falls at a much slower rate than in the linearized model. Now, if you put that to minus 15, you're really gonna see that the linearized model predicts deflation while the non-linear model doesn't predict deflation. Something that we have put the spotlight on in our earlier paper. Now, if you look at the other end of the curve, you'll see that as inflation goes up, so as inflation is in the area of say five, six, seven percent, you're gonna see differences between the non-linear and the linear model. In particular, you're gonna see that as the economy enters a boom phase, or if inflation is high, the economy is gonna see higher inflation rates going forward. Now, on the right-hand side, we redo the exact experiment for supply shocks. So these are shocks that move output and inflation in opposite directions, like for instance, the cost push shock. And you see that there is a similar feature embedded that it's a little harder to see than for demand shocks, but it's the same feature at work, namely that you're gonna get non-linearities in the Phillips curve. All right, so what do we do with this element now in terms of this, in this model? Well, we're gonna subject the model now to cost push shocks. So we're gonna subject it to increases in energy prices, to zero COVID, China, and our harbors, our lockdown strategies. So anything that sort of, say, makes firm's marginal cost go up in an exogenous way. And how do we do this in practice? Well, when we're gonna select given points on the x-axis, so when the output gap is at 5% or 10% or minus 5 and 10%, we're gonna subject the model to a one-time, once-than-a-deviation markup shock, positive or an adverse markup shock. And that's what you see in this graph. So the solid blue line is how does the Phillips curve shift in the linearized model if there's a once-than-a-deviation markup shock? And you see it changes, it's roughly half a percent. So the blue line, the Phillips curve shifts up linearly, and by the same amount, across the entire scatter region here. So that's something that, you know, the standards, Metz and Varus model in its linearized formulation would give us. We then redo exactly the same experiment for the nonlinear model, which has important implications. First, you see that, you know, the nonlinear Phillips curve shifts up too, but it shifts and tilts at the same time. If inflation is high to begin with, so if you are in the 10% output gap region and inflation is already ratcheted up to 8%, the same size markup shock pushes up inflation by a large amount compared to both the linearized model, but more importantly, compared to a situation when you are in an environment where inflation is subdued. So if you do the same experiment, say when the output gap is at minus 10 and inflation is below the central bank's target, the same markup shock produces only very little amplification. So in this sense, markup shocks are state, the effects of markup shocks are state dependent, dependent in our environment. So long story short, there is no such answer to what is the effect of a one standard deviation markup shock? Well, the answer is, it depends on the state of the economy. We redo this for supply shocks, and we're gonna get the same answer there, that if the economy is not driven by demand shocks, but if it's driven by supply shocks, we have the same implication there. So when the underlying economy is driven by markup shocks and you get additional markup shocks, you're gonna get the same amplification of cost push shocks. All right. Now what do we do with all this? Well, this is kind of keeping it still at the expository level in terms of demand and supply shocks. If, you know, for those of you who are familiar with the Smets and Vauders model, you know this model has seven different type of shocks. There's TFP shocks, there's financial shocks, there's government spending shocks, there's investment specific shocks, there's a whole battery of shocks. What we do in this paper is we then go through each shock at the time and redo the analysis that we've seen here. So what we do is we just select, say, the monetary policy shock and we simulate a long time series from the model for inflation and the output gap using a monetary policy shock. And then in each point in time, we're gonna add a little bit of a markup shock and we see how much amplification do we get there. And then we redo this for the financial shocks and for the government spending shocks, each at a time. What we find is that in all cases, for all the shocks in the model, we see that cost push shocks are amplified when inflation is high to begin with. So if inflation has reached 4% to 5% and we see an invasion of Russia into Ukraine and energy prices exploding, the non-linear model predicts a much larger kick on inflation than the linearized model does. So the left-hand subplot here shows that for monetary policy shocks, you see that the blue scatters is this is the effect of an additional markup shock on the rate of inflation depicted on the x-axis, on the y-axis. On the x-axis, we have the initial level of inflation. So this is something like a cheat sheet, say for policy makers or interested academics. If inflation initially is, say, 3%, then the linearized rise model predicts an additional markup shock is gonna lift inflation by about half a percent. And the nonlinear model gives you a similar answer. About half a percent. If you go to the right of the x-axis and if initial inflation, say, is 5%, so the economy has moved and has evolved and you're gonna start seeing inflation numbers at the level of 5%, you're gonna see now pronounced differences between the nonlinear and the linear model. In particular, you're gonna see that the nonlinear model suggests that the same markup shock has twice as a large effect on inflation than the linearized model. Rather than half a percent, you're gonna get 1%. Just in that little illustration. Now, if you're interested, we redo this for the other six shocks. So for financial shocks and demand shocks and government spending shocks, you're gonna get the same qualitative, sometimes even quantitative results. Each of these shocks has this amplification feature in. All right. Now, once you enter this nonlinear world on thinking about what drives inflation, especially at policy institutions, there's always this buzzword coming up. How much risk is there to the inflation outlook? How much inflation risk is embedded in the economy at the current juncture? When you solve this model nonlinearly, you're able to actually pin down how much risk is there in terms of inflation moving around. And what we show is that inflation risk depends on whether inflation is surging or whether inflation is descending. In particular, we show that inflation risk increases substantially when you are an app slope in terms of the rate of inflation already. So if inflation has started to creep up, this model, our model predicts that it's more likely that we're gonna see higher inflation going forward rather than when inflation is descending. The left plot makes that point. The red line is showing you what is the distribution of inflation if you are hit with adverse cost push shocks? What happens with the distribution of inflation when those cost push shocks hit you on an upward trajectory of inflation? So for instance, we have seen China, we have the supply side chain disruptions, we've seen all that, inflation was ratcheting up already and then comes the energy price shock on top of all that. In such an environment, you will see on average more probability mass shifted to higher inflation rates. Conversely, if you're on a downside slope already of inflation, you're gonna see less. And the non-linear model allows you to distinguish and work with that in a very rigorous way compared to the linearized model. All right. Now, with all this in mind, I wanna turn my last six minutes here on thinking about the implications for policymaking, particular monetary policymaking. And once you work with this model environment, what are the trade-offs for monetary policy makers? And I wanna make that using two different pieces of analysis. The first bullet, I'm gonna walk you through what happens if you look at impulse response functions. So we're gonna put the model into a position where the model replicates exactly the data in the first quarter of 2022. And then we subject the model, so the model really explains all the data in that very quarter exactly up to the fifth, sixth decimal. And then we subject the model to a cost push shock. So think about Ukraine or to a monetary policy shock. We're gonna look at how different these responses are to normal times. In the second part, we are then gonna look at the policy trade-off frontiers, namely how more complicated policymaking becomes for central bankers once inflation has left the bay and what needs to be done according to our analysis and how painful that can be in terms of the associated output losses. Now let's go for the impulse response functions. So what we do here is we are filtering the data with our non-linear model up to the first quarter of 2022. So the model exactly replicates that data. And then we're giving, we're kind of imposing a one standard deviation price cost push shock to that model. So think about the energy price explosion that took place. Of course here at a lower scale. What you see is that the solid blue line on the top left graph shows you the response of inflation. And so that our model predicts that inflation goes up by twice as much than the linear pendant type model does. So when you linearize the model, the answer would be that inflation goes up to one by 1%, the non-linear model predicts two, even two and a quarter of a percent. Note that that high in inflation does happen even though policy is much more aggressive in terms of how it reacts to policy in the non-linear model than in the linearized model. All right, so that's the effect of a price cost push in the first quarter of 2022. Let me turn to what happens if the central bank changes its course of action and deviates from the Taylor rule by hiking rates up and above its regular rule-based monetary policy rule. So in this plot, again, we are filtering the data up to the first quarter of 2022. And then we impose a one-time contractory monetary policy shock that top right subplot here increases the policy rate by 75 basis points on an annual basis. What you see is that monetary policy is more efficacious in containing inflation in the non-linear model. Inflation does fall by a quarter of a percent on an annual basis in the non-linear model compared to the linearized model. So good news is policy is more effective when inflation is high to begin with. Bad news is, once you see cost push shocks happening, that really ratchets us up the rate of inflation in non-linear terms, in the non-linear model. So it's a race between those two and it turns out that the cost push shock is stronger in terms of an inflation effect than the monetary policy effect is, especially when inflation is high to begin with. So if you're working in an environment where inflation is four to 5%, another cost push shock is really gonna take inflation off like a rocket and it's very hard for policy to contain that. Let me make that last point a little more concisely in the following way. I apologize for the little turbulent plot. I wanna study what happens if a central bank sees the cost push shock happening and wants to undo the effects of inflation. So think about an inflation not a central bank. So I see energy prices surging and I aspire as a governor to stabilize inflation at the level of inflation before the shock did happen. How much do I have to move my policy rate to achieve that? I want you to focus on the first column of this plot. The top left is the policy rate response over the first year in response to a markup shock or a cost push shock. Plotted as a function of the initial level of inflation. So if inflation initially is 2%, so if you are at 2% initially, a given cost push shock requires a central bank to raise its rate by about 2 percentage points per annum at an annual basis. So with 2% initial inflation, you have a moderate increase in inflation according to the model that actually coincides with the linear version of the model. Now, if you redo this analysis and inflation has crept up to say 6% or 8% and you have an energy price crisis, for instance, in Europe, you're gonna see that the required tightening is much, much stronger in the non-linear model. You would need to tighten policy much more aggressively and hawkish according to our model, namely by increasing the policy rate to about 6 to 8 percentage points compared to the no cost push shock scenario. Now, this is of course an extreme type policy. This is in the policy that keeps the rate of inflation constant, so that's the first column's second plot. That's a policy that aspires to exactly undo the energy price shock in terms of inflation. But it's a good benchmark to start thinking about maybe one of the polar cases. I'm gonna consider the second polar case next. Now, embedded, of course, in that is that you're gonna see a big fall in economic activity associated with stabilizing inflation that has great consequences, of course, for the output gap and economic activity on the labor market. You can go to the other polar case and that's what the second column shows you, namely what happens if you aspire to fully stabilize the output gap in term, when the markup shock hits. And so here the output gap is completely stabilized and you see that then changes in monetary policy rates aren't, so to say, big and most of the hit happens here on the rate of inflation. Importantly, again, the non-linear model predicts a bigger increase in inflation than the linearized model does. My time is up, so I'm gonna skip the remaining couple of slides, so what do these slides do? Well, these slides provide you with an assessment how does the model fit the post-COVID data. And we do this in two steps. We are computing forecasts. And these forecasts show two things. That the non-linear model, as inflation keeps on going up, becomes increasingly better at forecasting inflation. And more importantly, the non-linear model, very early on, as early as in the middle of 2001, 21s, offers a view that there's a risk that inflation may get out of hand very quickly. So the forecast with the forecast bands are much wider in the non-linear model compared to the linearized model, according to the non-linear model. So the non-linear model signals much, much earlier than the linearized model that there is a danger that inflation may get out of hand. Second, we show what are the shocks that you would need in both models to fit the data exactly, and you shouldn't be surprised if I tell you that the non-linear model requires smaller shocks to fit the data. And we're just gonna display that in terms of our analysis. All right, so let me wrap up and let me offer a few key takeaways. What I tried to do in the last half hour was to show you some of our research that we did in late 2021, early 2022, in working with a model that allows us to both think about the Great Recession and the vigorous increase in inflation rates post COVID in a unified framework. There's one model that does a good job in accounting for both of these very extreme cases of inflation dynamics in two very prominent economic episodes. Turns out the non-linear Philips Curse provide us a good way of thinking about this as a state dependence in the propagation of shocks and that turns out to be very helpful in thinking about, especially the post COVID period. And finally, the model offers rich implications for policymaking and I'd like to close with that and thank you very much again to the organizers for having me. It's been a big honor and pleasure to be here today. Thank you very much. Thank you, Matias. So I would now like to ask Anuka to the lectern to give the discussion. Thank you. Thank you. So I was very happy to get a chance to discuss this paper which I already knew quite well as we have also been very thankful for you to share the codes and be working on this paper somewhat already. Let me see if this works. Could you put the next slide? Okay, so Matias did a very good job presenting the paper already, so I will go quickly through some of the key features of the model. So essentially what this paper does is it shows that when you have a high inflation environment all the shocks essentially get amplified. The transmission gets amplified. So when you have a non-linear Philips Curve as a standard in the Smets and Bouters and you sold the model really non-linearly then you have quasi-king demand and which means that the impact of shocks is amplified in a high inflation environment. So the setting is a version of the Smets and Bouters model that they sold non-linearly or keep with the non-linear formulation and the parameters of the linear version are estimated mostly on the US data between 1965 and 2007 as Matias said, they exclude the great financial crisis periods as well as the pandemic and so on. And some parameters related to pricing and Kimball aggregator they are updated by calibrating or estimating separately to be consistent with micro-evidence and macro-estimations. So what is this quasi-king demand? So I copied these same graphs that Matias also had in his presentation. So it means that essentially the firms try to stabilize their markups or their profits. So when marginal costs are very low then the firms have very little incentive to cut prices because they can't create much demand. You can see that the demand elasticity in the top right panel is upward sloping and convex. So it means that if the firms were to cut prices in a situation where marginal costs are low consumers would not be willing to increase their consumption much because their demand is already high and satiated. The other opposite situation is when inflation is high and the marginal costs are high, we have energy shocks. Then in this case markups are low and the firms have to sort of stabilize that they get their marginal revenues up and even though they face a substantial drop in demand they would increase prices more than they would otherwise do. So what are the implications of having such a non-linear model? So in this graph we see that when inflation is high both demand and supply shocks get amplified suggesting that when you have a deeper downturn and higher inflation you will have that for a given supply shock. So essentially the Phillips curve becomes steeper and steeper in this situation. At the same time as Matias also alluded monetary policy transmission to inflation is amplified. So the impact of policy shock on inflation is actually larger. So you can stabilize inflation better and actually with a lower output cost as well because this really feeds into the inflation rather than the output. So the trade-off is lower but then once you take into account that these cost plus shocks propagate more strongly than the monetary policies shocks so when you put the two things together you actually have an increase in the trade-off. So I find the paper very, very interesting they have done a lot of exercises and really teased out the understanding of how all the shocks transmit in this model. So I really could not find much to comment in there. I went for the monetary policy exercise in here. I hope you can see these charts which Matias also showed just now that show that in the left side column you see in the left side column you see the responses of inflation or the forecast of inflation. You can compare the linear model in black and the non-linear model in blue and you can see that as you go further in time as you come closer and closer to the true crisis times and high inflation, the non-linear model then predicts the high inflation developments better. So I just picked up on one sentence that was in the paper which says that the focus are based on a quick normalization of policy rate which you see in the middle column. So you can see that the interest rate goes up to sort of six percent or so on in the focus. I'm not sure why this is jumping. The interest rate goes on to six percent or so on and the author is saying in the paper that if the rate path had been constant as in the data then the inflation forecast would be even higher. So then they draw the conclusion that our model suggests that loose monetary policy in the US is likely to have had an important role in driving the post-COVID inflation dynamic. So I wanted to analyze this a little bit if this could be indeed the case. So what I did was take the model as I had the codes and put it into the toolbox that also solves for optimal policy. And then I filtered the linear model. This is not able to handle non-linear models at least at the moment. So I put the model in there and I filtered it and you can see the red line in here in the left hand side that's the annualized interest rate and annualized interest. I don't know where this is. I'm not sure what is going on but this keeps jumping so maybe somebody is pressing the buttons a bit. Okay, so you see the annualized interest rate and it gets sort of similar result in this sort of forecast where the interest rate rises to around five percent or even above. And that suggests that policy indeed should have been higher according to this Taylor rule specification compared to the actual interest rate which you see in green in here. What was striking then I thought that perhaps the steady state interest rate is quite high and indeed it is so this is plotted in the solid red line in the left hand side graph and the steady state interest rate is 6.5 percent approximately in the model. That's because it's estimated up to two, data up to 2007. So this 6.5 percent level of the steady state interest rate could be compared to some measures that are out there. Maybe the more conservative estimate by Lubig and Mattis at 4.3 percent is the highest one that we have now. The survey of primary dealers puts the long run at 2.6 sort of consistent with Holstone Laubach and Williams at 2.5 and then Laubach and Williams at 3.2. So the 6.5 even compared to the conservative estimate of Lubig and Mattis seemed quite high. So what I did in this model is I set up a nut hook neutral rate in the model as a time varying intercept in the Taylor rule. I also tried a few other things like changing the beta or the sigma, the risk aversion parameter, but I thought this would be sort of a most cleanest ad hoc way of doing it. So I set up, you can see the equation here where I set the R star as an intercept in the Taylor rule and then I set up a process for it where it is AR1 process and just has a shock with AR1 parameter very high at 0.99. And then what I did was observe the neutral rate and I used the Lubig and Mattis measure. Now in this graph you see the red lines from the chart before as well as the black line which is now the new Taylor rule based interest rate path that tends to a much lower level so it doesn't peak anymore above 5 percent and it gets to a lower level at the end of 2025. So potentially that result that we have in the forecast of that interest rate path being very high is to a part driven by the very high steady state interest rate in the model. You can see the neutral rate path below which goes to around 4.3 percent. And the path, the black one is now more consistent maybe with the actual rate although it rises quicker than the actual rate which you see in green. Then I also checked what if the central bank was following optimal policies so I set up a simple loss function with equal weights on inflation and smoothing at one and the output gap weight on zero for now and because this model solves with expected shocks I set up the credibility of forward guidance to 0.3 so that forward guidance is dampened in a reasonable way and we don't have a forward guidance puzzle and you can find now you can see the blue line which is the optimal policy response and that is also quite close actually to the baseline response and it goes even to a lower level later on so that we don't have this undershooting in inflation at the end of the period. And this also is not too far off from the actual rate. Of course we have to keep in mind here that this is only done with the linearized model and if we had a non-linear model where inflation really gets much higher because of the amplification it would require even higher interest rate which would mean that probably these lines would be too low for them. Another thing that I noticed in the model is that the inflation target is set fairly high at 3.5% is also estimated because of course Fed only started doing inflation targeting at 2% fairly late. So in the model the target is 3.5 so then I wanted to study what is the consequence of lowering it to 2% and there are two things that are happening that are sort of countering each other. One is that of course the steady state nominal interest rate declined so that means that you don't have to raise the rates as much as you would have if the neutral rate is very high or the steady state interest rate is very high but at the same time policy has to bring inflation to a lower level so it needs to be tighter and these two effects sort of balance each other out and in this case you would have to have policy rate path which is actually quite similar to the one before we lowered the inflation target. In this case if you would do then optimal policy this would call for much higher interest rate as you see in blue here because the target is now lower so you have to bring inflation lower towards the end of the period and that's why you want to really have quite a bit higher interest rate in the optimal case. One could also consider then because all of these exercises were done with zero output gap weight so if we would set the output gap weight to 0.25 which would be consistent with equal weight across inflation, unemployment and smoothing by Okun's law we get sort of if we have one on unemployment we get by Okun's law to 0.25 on output gap and in this case now we have both the neutral rate in a model and we have a lower inflation target for all of that to be consistent. We get that the blue line, the optimal policies very close to the actual interest rate path in green but of course again this is only a linear model and potentially if we had the non-linear features here you would need to have a higher interest rate. Then in this case exercise I used a very conservative measure of Lupic amethyst for the nominal neutral rate which goes up to 4.3%. If I would use a measure which is somewhere in between the range of measures that I highlighted earlier, not yesterday then we would get a lower, that would be the Laubach and Williams measure which I think goes to 3.6 or so and I said the output gap weight to zero again and because the neutral rate is much lower you don't have to increase the interest rate so high. So in this case policy would be close to optimal again. If you look at the black line that's the Taylor rule based interest rate and that's much lower than the actual rate that has already developed and is expected to going forward and that would suggest that potentially even in a non-linear model, even in a non-linear model, the rate path that the Fed has followed could have been enough already and optimal policy then calls for something peak level of similar peak level as in the actual interest rate path but again slightly more earlier. That's it. So in my opinion the paper is excellent. I really enjoy reading it again. It's very inspiring how the authors explore the king demand and how they use the non-linear tools to tease out many exercises even though we know working with non-linear models is not easy because we cannot use the standard tools and using the inversion filter where you can invert the shocks and the variables the authors are able to then filter actually the model in the fully non-linear model and throw conclusions for the current period. And I understand that there's some merit estimating only until 2007 but in order to study the current period especially policy implications it will be important to look at some of the crucial parameters. Thank you. Thank you very much Anuka. And before I open up the floor for questions and I would also like to remind the online participants to file in their questions. I would like to give Matthias the opportunity to briefly respond to Anuka's discussion. Excellent. Thank you so much Anuka for a great presentation, a great discussion. I really appreciate. It's great food for thought for follow up work that we actually are pursuing as we speak. I like your simulations very much and I understand the desire to maybe rejigger some of the key parameters of the model beyond, so to say, and thinking about what's going on right now. Let me maybe give you a little bit of our perspective why we did what we did. We did not want to fiddle with parameters of this maths and values model all that much. And that was driven by, so to say, the idea okay let's do one change at a time which is isolate the effects of working with the non-linear model equations rather than the linear equations. And we were not, so to say, driven by a desire to change the steady state, normal interest rate, et cetera, et cetera. I agree with you that for further work, for follow up work one, so to say, auto look perhaps into these dimensions. And I think what you have done is a very interesting starting point for doing that. Understand it's a linear model based, so to say, implications. And extending that to the non-linear model that would be certainly super interesting. We had the position of, okay, let's not fiddle with the parameters and in fact let's give this model a hard time as possible to try to explain or come to terms with the data that we see. That was really, we didn't want to make this easy for the model. So that's why we froze the estimation or the parameters with data from 2007. That's a long time ago. But we really want to ask, okay, this model should have never seen the financial crisis, great recession, nor the rocket-shaped inflation takeoff. What, can this model get us any clue about these two events once we ask the model to tell us something about it? And so that was how we kind of motivated our kind of, disciplining our parameter choices. Now, again Anugai, I really appreciate your suggestions and as we speak, we have follow-up work going on and I would really appreciate taking a closer look at your work that you have done and seeing how we could perhaps sort of say address that in our subsequent work going forward. And so thank you so much, really appreciate it. Thank you, any questions? Yeah, Klaus, thank you. Yeah, Klaus Massen from the ECB. Wonderful discussion, wonderful paper. One, two questions, what explains in your model the increase in inflation before 22? No, I mean inflation in the US, in the other area was already quite high at that point. If Senate banks understand what you're showing, this non-linearity, and then if you already have high inflation, the cost per shock is even more difficult worse, then should Senate banks not react already to inflation which goes above 2% stronger than earlier before? That is the first question, why is, what is the setup of your model that you don't, at least in your presentation, I didn't hear that you discussed this episode. The 21, second half of 21, strong increase in inflation. What was the causes in your model and why did monetary policy not strongly react if they would know your model and results? Thank you. I suggest to collect, so it's fine with you. So over there was another one. Many thanks for the great talk and the great discussion. I was just wondering, the Phillips Cup can be convex either because the relationship between inflation and marginal cost is convex and that's what the Kimball demand cuts give you or because the relationship between marginal cost and demand is convex and stories like bottlenecks, et cetera would be more about this one. So question one, does it matter where the convexity comes from? And second, did you go for this one because you find it more plausible or for tractability purposes? Okay, and Giuseppe Ferrero from Bank of Italy, so we saw in Sintra, Francesco Lippi presenting a different state dependency in the Phillips Curve. It was about the frequency with which firms adjust prices. Here the state dependency is in the response to marginal cost. So I was wondering, since in that paper you have that when price increase increase more than what we have in linear model but also when they decrease, they decrease faster. What is the implication here if you can compare with those results? Thank you. And just behind you, Leo. You're from the ECB. Let me first say I enjoyed really very much the presentation and also discussion. So well done, very, very clear. The question I have is a very simple one. I mean, to me, and reflecting maybe a little bit my own ignorance. I mean, what you describe, Matthias, to me really looks like a very good comparison of what you can do in a linear or linearized environment in a non-linear environment going beyond this particular application. And I guess from an empirical perspective it can always be advantageous to go for the encompassing non-linear specification. Only maybe from a conceptual perspective when you want to build intuition, teaching purposes and so on, I see great advantages of sticking to a Mickey Mouse linearized version. So my question is, I mean, the only reason why I can imagine why, so to speak, we have not done what you did much earlier must come from other constraints, computational constraints, I guess. So what is your assessment? What is missing in terms of tools that nowadays are available from a computational perspective that on a routine basis we can do what you did and do not have to wait for another 15 years to wake up and say, ah, this was a space, obviously, a corner of the parameter space which we didn't have on our radar screens because we used linear methods. Thank you, so then I would hand back to you. Thank you very much for the many questions and I sense that's expressing interest which Jesper who is over there and I and of course our co-author Martin Haring really appreciate. Let me answer the questions in turn and let's go for the first question on what were the sources of the inflation uptick in 2021 now when we filter the data it's a mix of demand and supply shocks. The model has seven different structural shocks and it's a mix of demand and supply shocks and I cannot put my finger on exactly the ratio but it's about 50-50 if not geared a little more toward cost push shocks but these shocks are not extraordinary and these shocks are not super large. Now, your second question was, okay, so why didn't the Fed knowing what's to come according to our non-linear model? Why isn't it a non-linear model that policy was hiking rates further? Well, the devil's in the details and it's the Tata rule that is embedded in the model that did not dictate lift off yet, okay? So we did not in an attempt to not fiddle with parameters and policy specification it's the Tata rule that actually suggests a very shallow path. And exactly. Now where does this 3.5% inflation target come from? That's the average rate of inflation in the estimation sample since the mid-60s all the way to 2007, okay? Now, following Anuka's suggestions would be very interesting to look into what happens if we work with a lower inflation target and we digger some of the parameters pertaining to policy. I think the deeper issue is what did policy actually do and what should policy have done and what we are working on right now is on a project thinking about what are the optimal monetary policy prescriptions in such an environment. You wanna think about ahead, if inflation gets out of hand is it time to react more vigorously even though you may see transitory cost push shocks? So the notion of seeing through transitory shocks our hunch is that that becomes less of a weight if the central bank takes into account that if you're in an upswing of inflation you may wanna act a little more carefully in containing inflation toward the target without letting it go kind of all that quickly. Now, I really like your questions on where does the convexity come from and why did we choose Kimball? Now, we chose Kimball because we wanted to work with the Smets and Vowlers model and Kimball is part of that very model and it's a toolkit in many central banks and policy making institutions and our desire was to, you know, what happens if we wake up that so-to-say feature of the model and give it so-to-say the most important, give it all the spotlight it may deserve in case it gives us interesting implications. Now, you could put the convexities elsewhere you could think about firm-specific capital or input markets versus Kimball aggregation or put this more in the supplier on the demand side. Again, our basic desire was, okay, let's do one change at the time and work with the Kimball aggregator that Smets and Vowlers gives us. There's some very interesting work by David Lopez-Salido on co-authors before the financial crisis that actually looks into what are the implications of firm-specific factor markets and implied convexities, super interesting stuff. Now, again, our desire was to work with this Smets and Vowlers model and that could be very interesting work to consider going forward. I know that you have also done very interesting work on that, that, you know, Bridges that I'm looking forward to, you know, catch up over coffee. A very interesting comment made by, that in Cintra there was a paper apparently that documented that the frequency of price adjustments has changed recently so, quote unquote, prices become more flexible and in our model, that's not the case. In our model, the amount of price changes become larger but the underlying number of agility is the same. Now, does this matter or doesn't it matter? Well, it matters a great deal, especially for optimal policy making, right? If you are living in a world with high inflation where all of a sudden prices are essentially flexible, then disinflating becomes much less costly in terms of economic activity compared to a situation when prices remain, you know, sticky or there's more rigidity in price setting that has grave implications for the reduction required in economic activity. So again, we are focusing on that right now in some ongoing work thinking about the implications for monetary policy, you know, exactly along those lines, you know? So two observational equivalent frameworks, what are the implications for optimal policy, especially in the space for state dependent sacrifice ratios? Now, Leo, your take is very much welcome, namely, what constraints did we face or why hasn't this been done earlier? Well, I think in part it's driven by the availability of computational tools and knowledge that there is a tool out there that allows to calculate the non-linear solution of this model, extremely simple and extremely user-friendly, which is diner. There is a hidden gem inside diner that not many people know about. Most people working with diner go to the default solver, which is a perturbation-based solver for a second order, and that's it. And, you know, that's a tremendous progress for the profession in the last 20 years, for sure. But there's a hidden gem inside diner where you can actually, you know, just use a different solver flag and it solves the complete non-linear model under the assumption of certainly equivalent. Now, that solver is with us ever since 15 years ago, but not many people know about that, unfortunately. Computationally, this is not really a bottleneck. You can solve a SMETS-Vauders-type model with that solver in fractions of a second. I've been working with colleagues at the Board of Governors a number of years ago already where we solved models that had 1,000 equations and 1,000 unknowns, a multi-country model with that solver, and was able to solve it. So, long story short, we have the tools available, but it seems that some people haven't, there's part of the community of the profession that actually is not aware of these computational, so to say, facilities that are available at our fingertips, and so I can only urge you if you wanna know more about that. You know, I'm more than happy in the next two days to talk with you. I have codes on my website that actually provide you with example codes. I'm giving little courses once in a while here and there, teaching these methods to practitioners. Not that I wanna do an advertisement here, but these tools are available, Leo, and I can only encourage people to work with these tools going forward as my co-authors and I think these are important in thinking about issues that we face as macroeconomists these days. So, thank you so much again for all the comments, for the questions I'm looking forward to interact in the next two days with you. Thank you so much, and thank you, Anuka, again. Thank you so much. So, with that, I think we can move to the next paper which is on lose monetary policy and financial instability and it will be presented by Moritz Schulerich. Thank you. I guess the slides are gonna come in. Thank you so much for inviting us for being here. It's a great pleasure. Congratulations to the organizers for putting together such a fantastic program and apologies for kind of swapping the order of this. Those of you who live in a country with better train systems don't know this feeling when you arrive at the station at 6.30 in the morning, open your app and have a complex matrix of cancellations and delays in front of you and a complex optimization problem to solve at that moment in time. I actually did well and you should have seen this smile on my face when I sat in the car from the station and said, I'm gonna make it. I'm gonna be there 10 minutes before my session only to get stuck in security downstairs. So, this is my beginning of the day. So, I hope it's gonna be better for me here on. Anyway, so lose monetary policy and financial stability. This is joint work with Max Grimm, who is a graded student from Bonn-Lizzi's at the MIT right now, Oscar and Allen, an old team. And what we do in this paper is to ask a question that doesn't move, I'm afraid to say. Okay, here we go, no? Here we go. Now I can just click forward. Okay, great. So, we come back to a question that's been asked and is regularly discussed in the media and in policy circles. Namely the role of lose monetary policy in driving boom bust episodes in financial markets and triggering episodes of financial instability. It's been, lose monetary policy was blamed for the pre-2008 boom bust. It's been stressed again in the 2010s as a potential source of instability. You know the papers. And we have some theory that lose monetary policy incentivize high risk taking and increase in leverage in the financial sector. And there's also a very nice paper by Frédéric Bosse and co-authors showing how this might increase financial crisis risk. However, on the empirical side, things are not quite as satisfactory I think as we want them to be. There's a lot of, I should say this with a big caveat because there's one of the most important people in this literature is actually on this stage and it's gonna discuss my paper or paper and the one in me, José Luis. There is a lot of micro-evidence that links lose monetary policy, low interest rates to risk taking of banks to a general sort of risk, appetite, risk on episodes in financial markets. But if you look at the macro level, there is very little that links those two. And actually I was part of that author team that when I was at the New York Fed, we wrote this memo about monetary policy and financial stability. And we were looking for evidence linking monetary, the stance of monetary policy to financial instability sort of from a macro perspective and we couldn't find much. And this is the moment. So this is a very central bank paper that was triggered, the idea for this paper was triggered right then back when we wrote this memo. So the recent question is incredibly simple and it's almost like embarrassingly simple in the sense that we just ask, does a persistently lose stance of monetary policy increase the risk of financial instability? That's it. And most of the sort of value added, I think of the paper, is in operationizing this and getting the question onto the data and structuring and getting the research question in an empirical framework where we can address it. There is a note there that lose stance of monetary policy. What we mean by that is an interest rate set by the central bank. In this case it's gonna be a real interest rate that is below the R star, sort of the neutral rate. I'm gonna get a little bit more into that in just a second. So here's the argument and sort of the key theme and finding of the paper in a nutshell. Two charts on the left hand side, you see the year zero on the right of the horizontal axis is the year of the crisis and there's the red dotted line which is the US in 2007 and you see sort of in the years before the financial crisis there's essentially sort of the John Taylor et al argument that going into the 2007 global or US financial crisis, in this context I should say, interest rates in the US, real interest rates where the stance of monetary policy was very loose and real interest rates were fell below neutral. And then you also have the blue solid line together with confidence intervals. Those are the stance of monetary policy going into financial crisis for the universe of modern financial crisis going back all the way to 1870 for a sample of close to 20 industrial economies. So this is the core finding of what we have namely that there is a pattern in the data and I'll speak about this in the remainder of the talk that before a financial crisis the real interest rate tends to be depressed relative to the neutral rate or the stance of monetary policy tends to be persistently low and on the right hand side you get a hint of what we think is the mechanism that links the two namely the increase in credit growth and credit booms that are triggered by persistently loose stance of monetary policy. So that's the core argument and I'm now gonna walk through this summarizing again loose monetary policy is connected to medium term risk of financial instability loose monetary policy is connected to credit market overheating. So these were these two charts on the left hand side the medium term increase in the depressed rate before financial crisis and then on the right hand side the increase in typically the increase in credit growth and then we are gonna argue that we think we can and should think of these relationships as causal. I'm gonna have an IV strategy based on the macroeconomic trilemma so the idea that in under certain conditions of pegging your exchange rate under capital mobility your interest rates are set exogenously somewhere else and you have to import them we're gonna use that as an instrument and show that the loose stance of monetary policy here one percentage point looser increases roughly doubles crisis probabilities three years ahead. Okay, so what's the data? Data is from our own macros three database macro financial data banking crisis chronologies for 18 advanced economies since 1870. We also gonna use the bank equity crash and alternative crisis chronologies so the work that Matt Barron and Emil Werner and Wei Xiong have done, Rainer and Rogov et cetera the missing element this is all existing and this is all there out there the missing element for what we have in mind and what we want to do are estimates of our stock and I guess we're gonna talk about estimates of our stock of course there's some uncertainty around them it can't be observed so we're gonna define the stance of monetary policy which is gonna be our key impulse variable if you will is as the difference between the real rate and the natural rate of interest our stock which is the equilibrium real rate of return in the case of fully flexible prices to use Woodford's canonical definition. We're gonna operate in a very standard framework where we say that monetary policy affects the real economy through nominal rigidities and in the case where the real rate is below our stock we're gonna call monetary policy expansionary if the real rate is above our stock we're gonna call it restrictive. The definition of the stance that we're gonna use is this one so this is quite important so we're gonna have sort of a medium term of five year moving average of the difference between the real policy rate and the natural rate of interest. How are we gonna do our stock? How do we calculate our stock? We stand on the shoulders of a few former New York Fed colleagues namely we're gonna use the Delnegro et al. Method that they've done along they've calculated long run our stock for seven countries and we're gonna extend this to 18 countries so essentially it's a it's we're gonna identify country specific and global long run trends in our all connected in a VR model with common trends and long run restrictions on short term, long term interest rates, global interest rates and real rates and we're gonna extend what they do from seven to 18 countries so I'm gonna be reasonably short not only because it's quite technical but also because there's really not much value added from our side here. So this is what we get as a world trends in real interest this is the world trend in real interest rate and the red line is the seven country long run our star estimate by Delnegro and Tamba et al. The blue one is ours it's a little higher especially in the first in the sort of historical periods in the first half of the 20th century which is not surprising because relative to their dataset we add a lot of you know we a lot of Spain's and Portugals and countries were arguably maybe the real rates in this in earlier stages of catching up were higher. You see that for the past 30 or 40 years these two align very nicely. Now I'm not sure if I can click on these things probably not no so I would wanted to show you otherwise the country specific our stars as well as the time series of stands that result but the time super stands really is just we do this for all the 18 countries we have a long run our stuff for all the 18 countries and then the difference between that and the real policy rate is gonna be our stands measure. Okay, so the econometric model to start with is gonna be reasonably straightforward so we're gonna run essentially a local projections on an outcome variable B which is a positive event at a point in time and the positive event is gonna be a financial crisis or financial stability event either narratively coded in the JST way or quantitatively coded in the Baron Werner-Xiong way as an equity crash of the banking sector of the banking sector's capital base. These I'm gonna just tell you but these differences in the crisis definitions don't really make a difference. They hardly ever do make a difference and the overlap is very large. We're gonna then look at as our key variable is gonna be this stands variable and we're obviously interested in this coefficient beta H there for the various sort of projection periods. We're gonna have a vector of controls which has local and global control variables, anything from the typical macro was from growth and interest rates and inflation and we also have a bunch of country fixed effects. We, the global controls and time fixed effects give us very similar results so I'm gonna show you more parsimonious. We're gonna use these global controls and then we're gonna show you estimates of the beta H over a 12 year period so it's like it's a long period or 10 or 12 year horizon where we see if central banks keep policy rates or the real policy rate I should say low for an extended period below the natural rate we wanna see how this over the medium time effects the risk of financial instability going forward. And this is the key result. This is the key and so I'm gonna take one minute or two to run you through this. So there is a dashed line that you hope you can see in the back which is the unconditional crisis probability over three year period is roughly 10 12 between 10 and 11% and then you have the, we plot the beta H coefficient for the various periods and you get a shape of a time where a loose stands in the first over the first two to three years lowers the crisis risk relative to the unconditional crisis probability but then over after four to and then going up up to eight years out the financial crisis risk increases quite substantially if you look at the magnitude of the effect so we go from roughly 10 to around 14, 15% so it's not 100% obviously it's an increase in the probability that's quite marked and it's even taking the uncertainties of the estimation of R star into account these effects seem statistically quite clear and significant. There is, you see down there is a button of very loose MP so these effects gets much stronger if we look at periods where the top 20% of loose stands episodes for example we really have like where central banks kept the stands extremely given the historical sample obviously extremely loose for an extended period these crisis probabilities rise even further. So there's a lot of robustness and extensions let me just run you through the most important ones we, I already mentioned the different I already mentioned the different crisis chronologies that doesn't make a big difference it doesn't make a difference if we end the sample before the global financial crisis we can also instead of just doing this with crisis non-crisis we can do this with crisis with financial recessions versus normal recessions i.e. recessions linked to financial events if you vary into the sort of R star estimation we can do this with the Holston et al approach so using the sort of the sort of the New York Fed it's a version of R star we have, I mean I showed you linear probability model we can do this with a logistic model et cetera et cetera none of this changes this basic result and let me come back to this is essentially a loose stands of monetary policy ensures you against financial instability risk over like a short over the short horizon everything up to maybe four years maybe up to three years after year four we always see sort of this hump shaped curve that out five years out six, seven years out the financial crisis risk increases significantly so why does this happen why does excessively why does loose monetary policy trigger financial instability and I mentioned at the beginning in this slide on the right hand side that we hypothesize that the channel through which this happens is the financial sector is credit booms, is asset market overheating or the combination of both and obviously if we think this is the main channel then I owe you some test of that so we're gonna go through those we're gonna look at the role of this loose stands in triggering credit booms and triggering house price booms and especially their interaction if you know the recent Greenwood-Schleiferdall paper on these red zones so we're gonna look at this red zones where both the credit growth and asset prices are in a specific part of the distribution so we're gonna go one step back and ask does a loose stands of monetary policy trigger those dynamics so we will go through in the paper we go through all of those but let's do the red zone exercise because I think it's maybe the most credible one in the sense that this is a definition that Greenwood-Schleiferdall came up with so just to remind you it's being in the code as countries economies entering a red zone in year T if debt growth is over three year period is above the 18th percentile asset price growth is in the top third and the red zone then is a situation is a year where both the credit extension and the asset prices asset price growth is in these extremes or like in these top areas of the distribution or right hand side of the distribution and they differentiate as well between household red zones which is a combination of household credit growth and real estate prices in business sector our zones were business credit and real stock prices are in these parts of the distribution so on the next slide we're going to exactly go to repeat that Greenwood-Schleiferdall paper and look at the predictive content of our stands variable for these for entering red zones and it looks like this on left hand side the household sector on the right hand side the business sector it's clear for the household sector arguably you again have a dashed line there which is the unconditional Arizona probability and you see how a loose stands defined as before increases the probability of entering these household sector red zones i.e. joint credit and asset price boom situations and it's still there for the business sector it's a little the business sector lending data and the stock price data a little more noisier so it doesn't come out quite as clearly but we think it's still it's still quite there again if you do this with very loose monetary policy so you know the loose stands top 20th percent or above the 80th percent on the loose distribution then these effects get stronger again this is fine to do this not just for the Greenwood and I'll post World War II sample you can do this for all the years you can zoom in on you can add money growth and inflation you can run logistic models you can do decade fixed effects instead of the global controls this is all sort of this hump there's this shape that you know there is a the loose stands not immediately but over time triggers these joint asset price and credit dynamics is quite is quite strong in the data so none of this so far can make a good claim to causality obviously and so we're going to now propose an instrument for stands which is going to be an instrument that Alan Oscar and I have used in previous work and others have done too which builds on the so-called Trilemma of International Finance i.e. the fixed exchange rate under conditions of capital mobility you have to import monetary policy from abroad so we can for countries that run fixed exchange rate regimes under capital mobility build an instrument that just says that the variation in the local interest rate and in this case the variation in the local stands is affected by what happens abroad and you know the assumption is then I mean this is is that the the base country the policy setting in the base country doesn't pay all that much attention to the conditions in the home in the pegging country when setting rates so there's more to this we actually then also we also run a sort of tailor rule for the base country and just use the residual from that and sort of taking the expected changes in the base country out a few details are in the paper but the basic idea is simply that that you know if you think about the Scandinavian crisis in the early 90s the Bundesbank after unification raises interest rates the Swedes are packed and the Scandinavian currencies are packed so they have to raise their interest rates too into an asset price boom at the time and then you trigger the Scandinavian financial crisis I think that's these interest rate responses something that will hear a little bit in the discussion which I think is a very important point okay so we build this instrument so we and then instrument the stands with what comes from abroad and with the exogenous changes and monetary conditions imported to the pegging country here's where the ECB so this is a Eurozone perspective you get the stands in the 2008 Eurozone you get this idea that the core and the periphery had very different had very different stands for monetary policy the ECB I guess rightly focused mostly on what was happening and the big economies in the so-called core and that resulted in an import of monetary conditions arguably in Greece or Portugal or somewhere else that was to lose and this is sort of exactly the setting I mean just to sort of give you the idea of that instrument that we exploit in this case systematically over 150 years and you know for a long time most exchange rates were packed so we have quite a lot of observations there so we instrument the stands and then look at again come back to our first very first chart that I showed you the crisis risk in response to how monetary lose stands here then instrumented by the Trilemma IV effects crisis risk you see here these effects get considerably larger so relatively an unconditional crisis probability of around 10% we now have at a sort of 100 basis point lose the stands almost I mean I want to say triples this but like definitely doubles this again you get this very persist and this is the most persistent shape we get of all of these long run regressions is in the short run a lose stands isolates you or reduces crisis risk somewhere from year four or five on we see these probabilities increasing and as you see here if you buy into you know broadly speaking that there is some exogenous variation in monetary conditions that we pick up with this instrument then you will think that sort of causally then there is an increase in crisis risk down the road peaking somewhere around year seven, eight after the lose episode after the policy turns lose so if we instrument we do the same now again to think about the channel through which loose stands affects crisis risk again we think about mainly think about the role of credit conditions and asset price trends and here you see on the left hand side again the red zone Greenwood et al definition you can do but you can very much paint or see similar trends in just in credit growth or in asset prices deviations where similar way credit growth picks up then peaks around year eight that's nicely nicely in sync if you will or not so nice but in sync with the crisis risk probabilities and then then the probability red zone probability drops quite strongly I guess because there is a reversal and often in a reversal in a crisis again for the business sector things a little bit a little bit less clear that also has to do as I mentioned with the hide this is the common reminder this is a combination of having a credit growth to companies being in the top 20% and stock price growth being in the top third of the distribution of the country specific distribution and the stock prices are very noisy so things are a little bit less but I think that's something that is also in the original paper it's still paints the same picture okay so let me use my last few minutes to talk about something that I guess we as economists ultimately interested in I mean we care about financial stability I mean a little bit per se but also because or mainly because it affects the real economy so lose monetary lose stance might not be you know there's a there's a trade off in the air know there might not be bad per se at least we cannot say much about this trade off for now because we say like in the short run running this loose stance is expansionary that's why central banks choose a loose stance and that's the way of transmission and that's what we that's a way of stable stabilization that's what we that's why we do this but it might come at a cost and it might come might have sort of the the very loose have running a loose policy might come at the cost of financial instability in the medium run but also and is what we're going to test now it might come at growth costs at down the road so and we're going to test here is in line with you know some of the work that that that if me and or to be as Adrienne that others have done so in this whole growth at risk context is to look for negative medium real economic effect so do we see that a loose stance of monetary policy is also predictive of for a bad growth outcomes in the medium in the medium run you will this will remind you a little bit of the the me and Sufi a burner paper QGE paper with a service credit growth and then look at growth forward um this is not an access that we're not using credit growth but we're using the stance measure and ask sort of with the transmission through the credit markets if we see the this medium term medium term negative effect that the the downturn um being triggered by of being a being being a consequence of of a loose stance so what we're going to do is we're going to define low output growth as obviously for for for each country as output at being or growth changes being in the lowest fifth of the distribution uh we're also going to use baron or sewers series of economic disasters which they define as peak to trough faults in real GDP of at least 10 percent so we're going to replace the left hand side variable it's not going to be crisis risk it's not going to be our crisis probability it's not going to be red zone probability we're just going to say uh we're going to use instead low output growth or uh an economic disaster a variable from baron or sewers and this is how it looks um this is a growth risk trade-off emerging quite clearly from our from our work so again there's an unconditional low output growth probability which is 20 percent and then the um by construction and then the the loose stance um increases the probability of coming of an economy getting into a low output growth um environment uh by about what is it 20 20 percent relative to the baseline um maybe even more worrisome um if you think about avoiding the big disasters and and the welfare costs associated with them um is the result on the right hand side for the baron or sewer economic disasters so peak to trough slums in real GDP of 10 percent or more um here a loose monetary a loose policy stands again buys you some insurance in the in the empirical date and maybe i haven't stressed this enough but this is pretty much the universe of modern macro data that we have with the universe of modern crisis that we have so you know to overturn these shapes in these in these charts we'll have to wait quite a long time uh or at a lot of countries that for which we don't have data yet um but this this sort of hump shaped curve emerges very nicely again that the probability of disaster doesn't quite double but goes up by a significant amount after loose dance again it's not it's there's a there's a short run so benign um if you will uh benign or or or sort of insurance effect maybe and then um the disaster probability rises significantly um so um let me let me conclude and and maybe also finish a minute early um which is um which is um a uh this paper tried to answer that question for which in my reading of the literature when we wrote this memo for the for the for the board uh the literature didn't have an an answer namely what's the macro evidence that all this sort of risk-taking micro literature that has been well documented you know umpteenth papers about the bank risk-taking channel and how banks respond to low interest rates but if you i think if you if we're we're sort of we looked at the evidence and at the literature back then and say like yeah there's a lot of this micro stuff but we really don't know if any of this micro stuff adds to adds up to much in macro i mean i guess that's an endemic problem with some of the finance literature that people identify very well identify very well often effects but we don't really know how to think about them in a macro context and our conclusion at the time was that after reading a lot of micro papers that yes there are effects but we don't really know how to um what to tell monetary policy makers how important these effects are on the for the macro economy um and so we sat down gathered this data set and and and and and constructed these long run ASTA estimates using using macros and and and timbers et cetera methodology and now i think we we were in a position to say that um there is evidence you know and you open to you and then you decide how convincing you think that is in these broad cross country long run data that episodes of when monetary policy real policy rates were below the the neutral the equilibrium rate for this sort of you know remember we do this five-year moving average for an extended period this has measurable impact on crisis probabilities going forward the effects are first in the first short run things look quite good and maybe even buy some insurance against the crisis but from year five peaking around year seven eight crisis probabilities uh shoot um i mean shouldn't say shoot up but they increase significantly um we also showed you and i think that's uh that's the point that was important also to to us is given off pure research agenda that the evidence linking loose monetary policy stands to uh crisis risk running through credit and asset markets is there it's broadly compatible with what we uh what we find so this is you know we can at least open the black box a little bit and and and make sense of these developments and i think that might have nice interesting applications for macro proof frameworks et cetera that really we you know you can point to the from a macro perspective now also to the factors that intermediate uh from uh from loose policy or expansionary policy to uh crisis risk and uh good and then last but not least there are real effects and these real effects are potentially um uh uh severe and so they have to be valued against the short run gains of uh of loose or expansionary monetary policy i leave it here and look forward to the discussion thank you for your time thank you very much for the presentation for being on time both we're starting and ending and now the discussion will be done by José Luis Pedro yeah if i can have the slides uh and can i move the slides okay so let me start with um thanking the the organizers for uh inviting me to discuss these very nice papers i've been discussing Scholarich, Giordano and Taylor uh many other papers and i will be uh again discussing next month another paper so so it's always is also always always a great pleasure and also it's a great pleasure uh to be here let me just put one summary uh one page summary of the paper there is this motivation that commentators have argued that low monetary policy rates might create excessive restaking if you go back to the Jackson Hall paper uh of Ragu Rajan in 2005 and many many other people Jeremy Stein when he was governor of the of the Fed and there has been theory built up on that and there is a huge uh empirical evidence and this huge empirical evidence is more micro in the sense of low level data security level data etc so what is missing is what is in the black in in ball which is this evidence lack of evidence of aggregated banking crisis and this is what the paper of Scholarich and co-authors uh bring uh bring a new analysis now what do they do they use this so the question is whether the stance of monetary policy over a persistent period of time this increases the likelihood of uh banking crisis uh they use this I might say super interesting uh an amazing data of macro history uh uh you know by by Scholarich uh Jurda and Taylor they have also the banking chronology of crisis and they add in this paper these are a start uh building in another paper by Del Negro etc and this is the new data that they bring in and basically the results in the last part of the slides is basically that after a period after a long period of soft monetary policy uh this increases in a quantitatively sense the probability of a banking crisis and the intermediary channels are credit uh credit boons and asset prices and this will be the uh the the summary so I have one general command and two specific commands and I will develop the two specific commands on the general commands and not because I'm here and and I and I truly believe and I use this data these are truly excellent papers this in particular is a truly excellent paper both on on the methods on the question and on the interesting results and I I wouldn't stop here and I truly wouldn't stop here because that's that's the but of course the organizers invited me to say something more and so I'm gonna tell two specific comments the first one I'm gonna say is that it's not just purely low internal rates there is also higher monetary policy rates that might matter and I will argue with several slides why is this the case I will put motivating slides on why this is the case and then I will go a paper by uh two uh former colleague colleagues of mine Dimitri Kunishok and Bjorn Richter from Pompeo Fabra and Alon Tarquotho Gabriel Jiménez from Bank of Spain now our paper is complements I think complements perfectly the paper by Shulaik by Kreen et al and moreover and more importantly in order to say is we built on the paper of Yorda Shulaik Taylor on the amazing data and service to the whole profession that you can go google and get the data okay so let me put this there so this will be my first comment I will have a couple of slides on this the second one is gonna be on the R start and the R start you know if you read the book by Butfor that might be very important for many many and and of course I am here in a central bank it might be very important for the stance of monetary policy but I want to argue apart from some of issues on the estimation on the R start I want to bring something more related to whether for banks in particular for banking crisis I want to argue whether only these deviations from R start matters or for banks as we saw in Silicon Bailey bank but I will also put you other examples whether also nominal rates matter okay so these are gonna be my two my two comments let me start by the first one so in the first one these are important case studies of important financial crisis banking crisis I put since I am also now in London I put the variance crisis when they put out of money variance in Argentina Scandinavian crisis the US let me go from this US the 1929 the Japanese crisis and the Spanish crisis since I am from Barcelona so the first one is like as you can see the crisis started you know the banking crisis started in 1930 but the stock market prices started in 1929 which is here as you can see there were a lot of it is like a you of monetary policy rate the lower the interest rates they created a credit boom an unasset price boom in particular stock prices and then they raise monetary policy rates and the crisis caps this is not going to happen with recessions this is not going to happen with financial with non-financial recessions even with non-financial recessions deep recessions this is something only to do with banking crisis is the same story with the Japanese crisis okay so the Japanese with the platz agreement they lower they started increasing the interest rates in 1995 and until the start of the crisis in 1990 and my country Spain have a similar issue okay so these are some cases studies I will argue that these are not cases studies I didn't cherry pick these cases studies I'm going to show you that all these big big financial crisis are like this but before telling you something like this I want to use a person who is very important in the profession which is and I took it from a paper by David Lopez Salido which is Bernanke and Bernanke said if you read it here look at this part the market crash of October 1929 show that the concern effort by the Fed can bring down stock prices but the cost of this victory was very high you know you have a crash on the stock market prices and a banking crisis so these two the so far these two motivations are telling you that higher monetary policy rates are important not only low monetary policy rates and I might be a combination of the two and the last motivation evidence is from last two years so here I have many ones so in in 2022 both the Fed the ECB and the bank of England and other central bank started raising interest rates and we have seen problems in the Italian sovereign market the ECB even created the transmission protection instrument at that time yesterday in the financial times there were some issues in some countries including the United States especially but also Italy so sovereign dead markets you know I was living at that time and I was in the bank of England with the pension LDI funds and the guilds there were the crypto distress there has been other ones for instance the vice president here of the European Central Bank said recently that the commercial and real estate potentially has a problem and since we are more since this paper is more about banks let me just highlight on bank failures here when I highlight buying failures of the Silicon Valley bank etc I mean it's clear that you shape of monetary policy rates because these banks they took a lot of risk when the interest rates were low and when they created all these problems is when the interest rates went up okay so you know it seems in the last year that also not only in history but in the last year higher monetary policy rates in combination with previous low rates might have created a problem I have a paper which is related to Moritz which is this paper is I will use the same data JST means Giordà, Scholaric and Taylor I have this paper with Jiménez, Cunichoc and Richter what do we find in this paper it's highlighted in bold so monetary policy rates so low monetary policy rates create a banking crisis but only if only if you then raise strongly you raise you hide strongly the monetary policy rates this is for the last 150 years this is not last year this is 150 years using the the data by Scholaric and and co-authors so it's a combination of two and I will argue why is the case with with the channels one and this is not non-financial recessions not at all non-financial recessions not at all even deep non-financial recessions let me just put you here the post-world war deep crisis the worst ones all of them all of them came from a you you lower the monetary policy rates for a long period of time there was a credit boom and asset price boom and then when you raise significantly the monetary policy rates the crisis came not from the other ones and all of these monetary policy cycles have the same number of observations it's not that just you have one particular monetary policy cycle and that's why you get the crisis and this does not happen with non-financial recessions and not even with strong non-financial recessions what are the channels the channels Moritz was talking about the this red zone and in the red zone it is a paper by Greenwood and co-author Andreas Leifer in the Journal of Finance that basically they said if you are in a red zone that's the best predictor you have high credit you know high credit and high asset prices then with high probability high probability you will have a crisis and in fact they say lean against the wind and raise the monetary policy rates what we find in our paper is that not the case if you lean against the wind once you are in the once you are in this boom of the red the red per se the red zone per se does not translate into a financial crisis it is a combination of the red zone with this view of monetary policy because the view of monetary policy brings you the excessive risk taking in credit and asset pricing and when you raise the monetary policy rates you cut on asset prices you cut on credit and the crisis comes so in fact the leaning against the the wind by raising monetary policy rates in the red zone is too late that will be that will be this paper another thing i want to say i have three minutes is we would like to know a bit more on the channels and with these co-authors of mine we have apart from this amazing data that we built on Jorda Shulariq and Taylor we have the credit register of Spain that have a couple of crises and what we see there with low level data is that the defaults the low defaults come not just from low rates or high rates come from a combination of the two when you have low rates war war banks and risky borrowers they take a lot of risk that doesn't create defaults and then when you raise the monetary policy rates defaults are coming of course we don't have defaults yet which is a big mystery but there has been all these moratoria no in the last years with the covid and all these things but it seems that in 2023 defaults are going up and there are some perspectives no that in 2024 could be a big wars that will be the thing and let me so i have two minutes on the silicon belly bank let me just let me just say something on the silicon belly bank silicon belly bank many people thought okay let's only analyze public debt and in public debt automatically if uh interest rates go up asset prices go down and you have market losses from public debt and these are very visible because you have a mark to market but with loans is the same the thing is that you don't have a mark to market but if the loans are with variable interest rates and these people have to have have to pay higher interest rates they are going to default and in some countries they are going to have fixed rates but the maturity might be short might not be long and so they have to renew it and they would renew it at high interest rates and they will not be able to pay and if it was purely fixed rate which you don't renew it like in the case of marsukenberg i think no with the with the with the mortgage that he got in first republic i think then you have an interest rates there so basically something that we have not seen yet uh fully is the non-performing loans uh and that will be a channel and the last slide the last slide that i have uh and i have a 40 minute 45 seconds is you know like there are many we are in a central bank yesterday i was talking with a colleague of mine geordially there are people who are very macro economies and for them only matters the the the difference between the real rate minus the implied uh minus the arrestar now that will be perfect for macro models the boot for 2003 but in banking if you think about silicon belly bank think about silicon belly bank or the paper that i was putting it also matters nominal race and why matters nominal race is the same thing that we saw in silicon belly bank if you buy the internal rate if you buy the security at very low rate and you have to raise the internal rate even if it is not you are not changing the nominal or you are not changing the deviation uh with the arrestar you are going to have market losses and it's the same thing uh with loan defaults so and why is that because banking banking both the loans deposits and the securities that banks buy are a nominal race they are not a real race i'm not saying that the deviation i'm not saying more is that the deviations from arrestar are not important i'm just saying that for banks and this paper remember it's about bank crisis of banks it might also be important uh nominal uh nominal race and just the last thing uh Viral will present a paper it's a different paper but Viral's papers with Ragu Rajan and Stefan is a paper that is not just QE is just QE and QT and this is in a sense related it's not just low per se in this case expansive of the QE is just a combination for instance in that case QE QT here is a combination of low rates high rates thanks a lot thanks for invitation and it has been a pleasure to discuss your paper again before opening for the Q&A i would like to give more it's the opportunity to respond thank you yeah i mean thank you you were very generous thanks for these great comments i should mention that two of the co-authors are former students of mine Dmitri Kuzhnov and Birita so you came back with them and you know but it's it's um i think these i think these points are complementary i mean they complement complements in the sense that they uh they add to our knowledge so i i think what we can show what we try to show is that the um these sort of deviation the loose stands the deviation from us that triggers these credit dynamics and i think there is you know i think low rates no low nominal rates and the loose stands will be correlated to some extent i think that's i think that's fine and where our papers indeed silent is about what triggers the what triggers the crisis no and i think i'm very sympathetic to on the historical evidence that this is often the case when central banks turn the rate cycle uh when sometimes i mean sort of surprising leaning against surprises and leaning against the wind we have a paper on this tend to be very bad idea and and trigger crisis dynamics i think that's that's really really very much along these lines and and and absolutely absolutely true um i think we don't understand quite i mean maybe it would be a question i have back for you is that increase in rates is that is that a policy decision or is that sort of in a weak salient sense is that competition for market for funding and sort of rates um rates in the in the in the in the in the banking sector just uh go up because you run out of funds to fuel the credit boom etc so i think there's lots of open questions i think i'm very sympathetic overall to these um to these comments especially what is the trigger that then gets us from the credit as the price boom into the crisis mode and rates nominal rates in central bank actions likely probably go any questions from the room uh yeah so you will get a microphone thank you so this is a great paper uh big fan of the literature now that you're here i want to just try to uh to see if you can help me with something that i don't quite follow from these uh uh kind of research line so i i understood but uh that the people try to use these uh you know dynamic correlation to think about mechanisms so what is your traditional mechanism what what do i think about you know special features of the economy and try to tease out different way of thinking about the nature of crisis and why crisis appear and what is one of the key mechanisms think about theory and try to tease out element of theory but but you keep emphasizing the most of the most of the job with your effort is to try to identify a causal effect so try to find try to connect policy actions to you know outcomes in particular you try to lean policy actions to uh crisis outcomes that's a that's a fantastic uh research question so but but most of the analysis that you do was pretty much what i call in sample prediction right so you have a sample and you try to predict you know within the sample certain crisis certain type of events uncorrelated those events with a measure what you call policy have you tried to think about the extent to which you can do some uh out of sample uh analysis so i mean Jose Luis was alluding to something like that i would be interesting to me to see uh whether you can run some sort of uh you know uh out of sample posterior analysis to try to pick up few exercises to few few countries uh in which you know with your model try to understand whether the prediction that you are claiming are causal are very helpful in actually peeing down you know these these crisis probabilities and it could be you know you know situation that satisfies some of the criteria that you have to identify what is the exogenous component of these you know way of measuring the policy stance so it will be interesting to see whether you can show these with these beautiful data set that you'll be building over time so this is great uh thank you so much thank you yeah i'd like to collect so color was here okay thanks carlo altavilla ECB now uh just there in the spirit of the conference that would bridge practice in science do you have probably mind an average cycle that would characterize your results so monetary policy lose for but how lose how much is how it's loose so the size the depth of this loosening the persistence how much is persistent so if you can maybe describe probably what is in the probably what comes out of the result and second would be whether there are a heterogeneity in the cross-section of country that would more would more influence the probability of the crisis so probably that is a part there for sure there are policy mistakes in the crisis in the in the in the sample or there is a huge deviation in GDP or in the other star maybe all these also play a large role thanks very nice work i enjoyed a lot i want to ask you know i've never heard this concept of red zone which was cool i thought but i wanted to ask you more is have they sort of following up with david said have they sort of checked this sort of out of sample have they done some you know when they it seemed like you're really picking on certain person types that make you suspicious that it's not sort of robust but if you could tell us a little bit how you sort of check that this is a reliable measure and then i was curious one of the first figures you showed with the sort of the persistent decline in policy rate r minus r star say persistent two percent would give you say 10 percent on the credit to gdp it doesn't strike me a lot no if i think about sort of you know in sweden you know the run up when the ricks bank you know had this discussion where they were leaning against when you had credit gdp running around flee 180 percent of gdp you know are you going to trade that off you know 10 percent down say say 108 instead of 190 for persistent two percent hike in the policy rate that's that's like a tough tough trade off there just want to hear your thoughts about that thanks maybe one or two more short ones no hundred percent yeah clausen fascinating discussion the author and the discussion could go together and and ask the following try to answer the following question if monitor policies too is very expansionary for a long period it could show does it only show up in asset prices does it show up in inflation and if it show up in inflation is a reaction roce louis showed a reaction to higher inflation and if my policy that would also be interesting does not react because there may be a financial dominance situation they let inflation go to avoid the financial crisis so you have two additional trade-offs when you look at the full spectrum of what the expansionary policy would lead to in terms of inflation and how then monitor policies react knowing the trade-off between letting inflation go or creating a financial crisis thanks so then the final one would be there thanks maritz a real really great paper really just one question related to roose louis point but sticking to real rate because it seems that some of the channel you emphasize on house prices credit growth etc would apply if interest rate are low just because our star is low and not because r is below our star so can you say more about why use r minus r star and not just r and have you done it with r and does it give similar results that point thank you so much i guess the last one we've tried r star it's really about the gap that we get most of the action so the low r star itself in our framework is not to the same extent you get similar shapes but like the the the sort of the significance and and these and the the precision of in the estimates really comes from the stance variable same for inflation i'm going to be short and you can answer the difficult part of that we see a little bit of that i mean you know in the boom inflation goes up so that could be the trade-off that you described that could be could be real and you know we might want to look at the outcomes with these credit booms that were not followed by rate hikes to what extent that is these trade-offs are materialized in in the data and we can study them but to these comments here that which are excellent i think yeah absolutely i do agree this is not a this was not designed as a policy exercise this was we started off it like looking looking at the data let's see what's out there in terms of as you're saying in sample analysis of what's there i mean we what i showed you we cut the sample we split the sample we left out of 2008 crisis and these relationships remain there but it's not the same as saying with real-time information available to policy makers at the time um is these are these predictive uh is this predictive content there that would be an amazing exercise it would be high it would be i think it's a separate paper would be an amazing exercise i'd be we we've done a paper with björn actually and another co-author where we did something like do in real time if you have asset price and credit data again the non-revised but the original that you know all these macro data get revised all the time so it really doing this for 150 years is it would be tremendously valuable but it's it's a very difficult very difficult exercise in this paper um with björn and and povak tell actually we found that in real time sort of with the data available to policy makers in real time not using any look ahead bias just looking looking sort of one sided filtering or whatever you want to do um there still is predictive information in but this is not this paper it's a different paper in in credit and and asset prices for financial crisis but the aster we haven't done i think that's that complicates the whole exercise but the trade-offs and and and the swedish example i think laris always comes up with the swedish example as well it is true um that as some of these um so velocities you will don't look so attractive uh absolutely true um on the other hand um 10 of gdp is also not nothing and you know if you think about what might really be the sort of the tip of the iceberg or that what really triggers the the the x of speculation maybe that is the 10 where we want to avoid but i agree like sort of the doing the back of the envelope calculation it's it's it's it's not it doesn't look very attractive yeah so before before i go to that on the red zone is is particular for both it's particular on a very short period of time we took the red zone independently because we don't cherry pick the red zone and it was in a particular point in time and it works for the 150 years and in fact if you switch if you cherry pick a small thing you get a united states because united states in the you know in the 1929 crisis doesn't enter because the red zone there it was almost a red zone but it was not a red zone okay so this is out of sample estimation and then on the inflation coming from Spain there is another thing that it doesn't tackle in this paper which is just credit aggregate and asset prices both you know for business and for real estate but in in in in Spain in the boom there was a huge you know non-tradable you know boom which was i think inflation was important because the tradeable could not compete with the high inflation from the non-tradable so this is something that we both we could both of papers could incorporate and would answer to your question thank you okay thank you very much to everybody so this concludes the first session thank you to the presenters discussions and also to you the audience with the questions we now will have lunch so lunch will be served in the four years so if you go through the door and pass on to the back and i would like to ask you to be back on time at 2 pm when the next session will start thank you very much