 So, what do we do in this paper? We look how financial markets react to monetary policy announcements in the first half hour. Then we separate the effect of news about monetary policy in the announcement and news about the economy. Then we track how the economy responds. We do it for the U.S. and for the Euro area and we use a vector out of aggression. What we find is the following. First, market reactions reflect both news about monetary policy and news about the economy. The latter account for non-trivial share of the variance is much as 35% in the U.S. and even 45% in the Euro area. Depending on the mix of news in the announcement, the response of the economy is very different. We find that surprise interest rate increases are contractionary when they reflect news about monetary policies or when they reflect a monetary policy shock. However, they can also be expansionary and this happens when they reflect news about the economy. So, the main implications are two-fold. First, private agents learn something about the economy, not just about monetary policy from central bank announcements. And second, these news about the economy, they attenuate the standard estimates of monetary policy effects because we mix expansionary with contractionary shocks. So, the plan of the presentation is the following. I'm going to start by showing you some data. Then I'll explain how we relate to selected literature and then I'll give you more details about the VR, the identification, the impulse responses that we find and finally quickly explain the structural DSG interpretation. So the key data in this analysis are what's called the surprises. So if P is a price of a financial asset and tau is the time of the central bank announcement, surprise is the change in the financial asset between 10 minutes before and 20 minutes after the announcement. So we compute the surprises for interest rate derivatives, for Fed Fund's futures in the U.S. and for the Ionia swaps in the Eur area. So our motivation is to look at the instrument that's liquid and that reflects not just current policy rates but also their expectations sometime into the future. So that reflects some near-term forward guidance. So we compute the surprises for those interest rate instruments and for the stock prices, the S&P 500 in the U.S. and the Euro stocks 50 for the Eur area. So to make it very concrete, let's look at one example of an FOMC announcement. So this one comes on March 20, 2001 at 2.15 p.m. So the following text is released to the public. The Federal Open Market Committee at its meeting today decided to lower its target for the federal funds rate by 50 basis points to 5%. What follows are a few paragraphs of the analysis which I have clipped, but the bottom line is that this analysis suggests substantial risks that demand and production could remain soft. Now we know that this decision to cut Fed Fund's rate by 50 basis points surprised the market. It was not priced in. If it had been priced in, we wouldn't see nothing in the prices of the Fed Fund futures. However, we know that Fed Fund futures dropped between 205 and 235, our window for computing the surprises. So the markets did not price in such a big cut. They were expecting some interest rate, some policy easing, but not as large as 50 basis points. So this was unambiguously a surprisingly easy monetary policy. Now what's interesting is that stock prices also dropped in the same half hour window. And this is surprising in light of a simple textbook analysis which tells us that stock prices and interest rates should move negatively, conditional monetary policy shocks. So if the Fed lowers the Fed Fund's rate, this stimulates the economy, this makes us expect higher future dividends, we discount those future dividends with a lower interest rate. So the present discounted value of future dividends unambiguously goes up, meaning that stock prices should go up. However, they did not go up on March 20, 2001. And more generally, they do not always go up after surprise interest rate cuts, policy rates cuts. And this is the message of this picture. So here, every dot in the scatter plot, every dot is one FOMC announcement. On the horizontal axis, you see how the Fed Fund futures changed in a half hour window and containing the announcement. And the vertical axis shows how much stock prices, S&P 500, changed in the same window. So the theory just outlined predicts that these points should be aligned along a negatively sloped line through the origin. And what we see is that such a negatively sloped line through the origin is a fairly poor description of the picture that we see. So the numbers in the corners, they count the interior points in different quadrants of the plot. So we see that about one third of the points are in the wrong quadrants. Now in the quadrants one and three, where stock prices and interest rates can move positively and not negatively. And also looking at the right quadrant, we see that some large interest rate cuts are accompanied by very small increases in the stock prices or some small interest rate cuts are accompanied by large increases in stock prices. So the relationship is weak. And so there are different reasons why it could be weak, why it could be noise, measurement error. But the story that we are exploring in this paper is that it's the information about the economy in the announcement that moves Fed Fund futures and stock prices independently of the information about monetary policy. So specifically, we think that those points are manifestations of two underlying shocks. A monetary policy shock, which induces this negative movement between interest rates and stock prices. And some other shock, which we call a central bank information shock, which generates a positive movement. So now in the rest of the paper, we investigate whether, depending on the mix of the two shocks, the response of the economy differ. We use, for that we use a structure of ER with a mix of sign restrictions and high-frequency identification. So that's what the paper is about. Now let me say how we relate to some selected literature. First there is quite a large literature, which is here represented by just a few citations, that studies interest rate surprises as proxies for monetary policy shocks. And then uses those proxies to estimate the effects of monetary policy. So what we add to this large literature is adding the information shock. Then there is a smaller and more recent subset of that literature, which also looks at central bank information. The information about the economy contained in the announcements. And they follow mainly two approaches. One is to proxy the Fed's private information by looking at the differences between Fed's internal and confidential forecasts and the market forecasts at the same time. Another strand of this literature looks at the announcements themselves and uses text processing tools to process this and interpret these announcements. So we don't do any of this. We outsource it to the markets to read these announcements for us and to appropriately digest them and to react according to what's surprising in them and to what extent they contain surprises. So we've recently learned about these three papers that also follow this broadly, this approach. So they also use market reactions to tease out the information shocks from the announcements. But they differ in various implementation details and in the scope. So finally, I want to mention models of the information channel of monetary policy by Nakamura and Steisen and Melosi. So these are models where there's some uncertainty about the economy and agents figure out from interest rate decisions what the central bank thinks about the economy and this helps them sharpen their own views. But what we add to this is we add an independent communication policy. So agents learn not just from interest rate decisions but also from what the central bank talks. How it talks about the economy. Moreover, we use a VR. So let's get to the details of this VR. So I'll focus on the baseline model that we run for the US. So we have a vector of seven variables. The first two are the interest rate surprise and the stock price surprise. We map those half hour, those intraday surprises onto a monthly indicator because our VR will be monthly. So we construct a monthly indicator that collects any surprises that occurred in a given month. The source is well-known REFET's data set. And then the following five variables are the standard macro and financial variables such as government bond yields, stock prices, but this time the monthly average of them, GDP and GDP deflator and the excess bond premium as a proxy for financial conditions. So these are the variables in our VR and we run a VR. Now the only thing that's unusual about this VR is that the first two variables, they're IAD, right? They are surprises. So by definition, they're not predictable from the past of anything else. And therefore, this motivates the zero restrictions in the VR. Other than that, the VR is standard and we estimate it with standard Bayesian methods with standard priors and so the GIP sampler with which you obtain the posterior also allows us to impute the missing values of M that sometimes occur. So now this is the crucial slide and it shows how we identify shocks in this VR. So we identified just two shocks. So this table shows you the restrictions on the impact impulse responses that we're imposing on all variables and for these two shocks. So the first two variables are the surprises and the zero restrictions reflect the fact that these surprises are measured in a narrow half hour window and therefore they're unlikely to systematically represent all other shocks that are happening in the economy, right? Now the sign restrictions reflect our assumptions about the commovement conditionally on both shocks that we investigate. So notice that all our restrictions are on this high frequency part of the VR. We put no restrictions whatsoever on the low frequency variables on the Ys. Dot means no restriction. So this decomposes the surprises into two orthogonal shocks and what we impose are just signs of the commovement, right? So this is a set identification. This is not a sharp identification, which means that there is an uncertainty about the rotation. We're indifferent about any rotations of the two orthogonal shocks that are still consistent with the sign restrictions and this inflates our error bands appropriately. Now as a robustness check, we do something much simpler than that. Namely, it's something that for a lack of better idea, we call the poor man's sign restriction. So in this procedure, we just manually classify events as either policy shocks or information shocks based on the commovement. So whenever the commovement is negative, we classify it as a monetary policy shock. When it's positive, we classify it as an information shock. And the results are similar in either case. So now for comparison, let's consider the standard high frequency identification in which one just uses a single surprise, an interest rate surprise. And this single interest rate surprise is considered as a proxy for a monetary policy shock. So here we abstract from information effects like most of the literature and we just use the VR to track how the economy responds to these interest rate surprises. So let me now start with the results and let's actually start with the standard high frequency identification. So these are the responses of the variables in our VR to one standard deviation positive interest rate surprise. So in the first row, you see just a blip because the surprises are just blips, they're IID. So that's by construction. And then in the following rows, you see how the low frequency variables in the VR respond. So you see a protracted increase in the bond yield, decline in the stock market and gradual effect on GDP and prices and some increase indexes bond premium. So in particular, I want you to notice this very gradual response of the price level, which is a typical finding in the VR literature, right? So in many VRs, we find the price puzzle, so actually a slightly positive response of prices. Initially, it becomes negative only after some time. So in this case, when you use this high frequency identification, you don't get a price puzzle. But still the response of the price level is extremely slow. So now let's run our more sophisticated VR that distinguishes these two shocks, mantra policy shock and central bank information shock. And then the first two rows, this time we need two surprises to uncover these two shocks. And the first two rows, you again see blips and moreover these blips, they're not surprising because they just reflect our identifying assumptions. So let me now just zoom in on the low frequency variables that are unrestricted and therefore all the results are sitting here. So you see that this sign restriction that we impose, the restriction that we impose on the commovement of the interest rate and stock price surprises, that this simple restriction separates two completely different shocks, a contractionary shock from an expansionary shock. So both shocks in both the second and third column, you see that both shocks are associated with an increase in the bond yield, but here the similarities end. So all the remaining impulse response are exactly opposite. So after a mantra policy shock, you see a contraction and after an information shock, you see an expansion. Now, and if you compare the, for example, again the GDP deflator, you see that the response of it is much more vigorous than in the case where we just use standard high frequency identification. So in light of our analysis, this very gradual response of prices is just an effect of a mix of surprises that cause a pretty vigorous response of prices and other types of surprises that of central bank surprises that cause an increase in the prices. And after you mix the two, you get this very, very gradual decline because in the end, mantra policy shocks dominate in this mix. So this is the core result of the paper. Now, this is what the shocks look over time. So this is just to illustrate that in our procedure every month, you see some different mix of mantra policy and central bank information shocks. They, both type of shocks occur throughout the sample. So it's not driven by some selected episodes. Now, we've repeated the same procedure for the Euro area. So first, we've created a data set of ECB announcement surprises using a similar approach. We have 284 ECB policy announcements from 1999 to 2016 and we cover both press releases and press conferences where also a lot of information is released and press conferences have been a standard feature of communication of the ECB since the start. And so when you look at this plot, this is the analogous scatter plot for the ECB surprises. When you look at this plot, you see that negatively sloped lines through the origin is even poorer description of what's happening here. So if anything, the mix, there is even more even mix between mantra policy shocks and information shocks. So then VR is, when we use the same, run the same VR, the results are similar to the US with this exception that they reflect that the share of information shocks in the mix is a bit higher in the Euro area. So if you look at the responses to just all interest rate surprises in this standard high frequency identification, you'll see that on average, positive interest rate surprises are followed by an increase in the S&P 500 and by a decline in the bond spread. However, when we separate the two shocks, they're based on the commovement of interest rates and stock prices, yep, you see that these results are a mix of a contractionary shocks and expansionary shocks. So now what does this imply for a standard calibration of standard DSG models? Now, so we look at these results through the lens of a DSG model at Gertler Karadi and New Keynesian model with financial frictions and we match their impulse responses. So when we match the impulse responses obtained with this high frequency identification, then remember this very gradual response of prices, we need very high nominal rigidities to match that. With these high nominal rigidities, we match the GDP decline without the need for any financial frictions. However, when we then match the impulse responses to the monitor pulse shocks that we estimate, that's clean of this information effects, we find that, well, you remember that prices respond faster, so you need less nominal rigidities to explain that. But then to match the responses of GDP, you need financial frictions in the picture, but that's great because this helps us to match this increase in the bonds price. So everything falls into place. So according to our identification, the nominal frictions are less important and financial frictions are more important than you would think looking at the standard high frequency identification. Now, when it comes to responses to the information shock, so these are more of an interesting, well, more of a non-trivial thing, because they're kind of new. So there are two possible stories that one can tell about these impulse responses. So one is that central banks have superior information about the fundamentals, and this is the, well, the story that goes back to Romer-Romer 2000, it has not been uncontested, but okay. So if the central banks have superior knowledge about fundamentals, the impulse response that we see is just a materialization, right? So the central bank sees some fundamentals better and their effects materialize in these impulse responses. So the central bank just predicts these responses. It does not cause them. The fundamentals are there, right? And the causes for the expansion of the economy are there anyway, and the agents would have figured it out anyway, right? That the boom is coming. So the only effect of the announcement is that agents learn a little earlier about those fundamentals, but they would have learned about them soon anyway, right? So there's very little causal effect of the announcement. The most of that is just that the central bank predicts the subsequent trajectory of the economy. This is the story of Nakamura and Steinstein, and we also have a story like this in the paper. But now there is an alternative story which is unexplored in the paper, which is that these announcements are self-fulfilling. So if the state of the economy, if the equilibrium depends on the level of confidence of the agents, or if there are stronger strategic complementarities, then a public signal does not need to be precise in order to affect the equilibrium strongly, right? So in this case, the announcement would cause the trajectories of the economy that we observe. So to conclude, we partition interest rate surprises into two components, a monetary policy shock, which is an interest rate increase which is followed by a contraction, and a central bank information shock which is followed by an expansion. The lessons, first, the effects of monetary policy on the economy are stronger than if you ignore these information effects, and the second, central bank information is relevant. However, we don't know if it causes or if it merely predicts the trajectory of the economy. Thank you. Thank you very much, Marek. We're discussing this effort. Thank you, and until my presentation goes up there, let me give you the basic roadmap. My presentation is going to have two subheadings. One is giving a history of thought. The second one is whining about the paper. I usually start from the second one and only do that, but given the mixed audience here, I thought it would be useful to talk about where this paper stands and our understanding of monetary economics and its measurement. Some of you might be surprised that in this day and age, we are still thinking whether monetary policy has real effects and how much and how soon and how large and how persistent, and that's because essentially, because as citizens we're lucky, but as economists we're all lucky that central banks don't do random policy actions. So in general, monetary policy changes for a reason and therefore the cause is that reason, not monetary policy, and therefore it's not easy to identify what contribution monetary policy has. So usually, I'm going back at least 40 years, we use VARs with quarterly data because GDP is measured quarterly and we care about that. And those VARs would have something like, let's say GDP and inflation and monetary policy, some interest rates. And the issue there is how do you identify the exogenous monetary policy innovation? The standard is the Cholesky factoring, that's essentially a bunch of zero restrictions and it says, okay, that which is affected by other macro developments this quarter but does not affect them is the monetary policy surprise. And that thing is called a shock. And that language is important and monetary policy shock is a orthogonal innovation in the policy and the orthogonality condition is with respect to the state of the economy. It's orthogonal to GDP and inflation. Good, and then you can trace out what that shock does to other stuff. We could think of other restrictions to identify, right? We can think of sign restrictions, we can think of long-run restrictions, we can think of other zero restrictions, there are many. But they all have the same idea, place enough restrictions such that you're gonna identify something that's orthogonal to the current state of the economy and you can tease out this effect on variables of interest. Now this leads to various puzzles, the best known is what Maric already talked about is the price puzzles, which is in a VAR of this sort, right? If you have a monetary policy shock today, inflation first seems to go up. And we kind of understand why. That's because your VAR is too small, your central bank actually knows more than what is in the VAR. These guys are in stupid, they see that inflation is coming, they raise interest rates, but your VAR doesn't capture this. So it seems like higher interest rates are causing higher inflation initially, okay? So, and this is something important. This is the beginning of the information problem. If the central bank knows more, then I'm gonna see this in the VAR in weird ways. Okay, good. Now, a kind of parallel literature is to say, okay, we give up on this high frequency, orthogonal shocks to the state of the economy altogether and we're gonna look at things that are very high frequency, where identification is trivial. I'm gonna look at what happens to the, let's say the 10 year yield, in the 10 minutes around the announcement of the policy. Their causality is clear. Policy does not react to what has happened to the 10 year yield five minutes ago. If there is a systematic relationship, clearly the causality goes from policy to your high frequency, other variables. Good. This goes back to work of Cook and Hahn and others, but the good work begins with the paper by Ken Kotner who says we can measure monetary policy surprises from market based indicators and use these precisely measured surprises as controls for the effect of monetary policy on other asset prices. Good. You do that and what you measure this time is called a surprise. That too is an orthogonality condition, but this time the condition is that the innovation is orthogonal to the information set of the market participants. These guys have learned something that they didn't know before. What they learned is, ooh, monetary policy tightened by 25 basis points, I thought they wouldn't change it now. Now these two things, the VAR shock and the event study surprise will be the same under certain conditions. In particular, if the market participants know exactly what is in the VAR, if those information sets overlap, then what is a shock for the VAR is the same thing as what is a surprise for the event study. If they differ, then they measure different animals. Cool. All right, now what is nice here is we no longer haggle about identification, however, we don't know how to move from here to not that I measured the surprise since this is not a VAR shock, I don't know how to think of how this relates to inflation six quarters from now. All right, it is also the case that if you run these high frequency regressions, you're gonna find much higher R squareds for fixed income stuff than for stock prices. And that's something very relevant for this paper. And the standard line, I'm a big proponent of this, is who understands stock prices anyway and monetary policy has multiple effects because on the one hand, you're jacking up interest rates, bad for stock prices. On the other hand, you're probably jacking up interest rates for a reason if the stock market participants think that, ooh, the central bank thinks things will be better. That's why they are raising interest rates. Then this is higher expected dividends in your near future. So these two effects pull in opposite directions. And in fact, this is the main identification mechanism in this paper. All right, now in between, if you run these regressions and if you haven't done this, you're gonna be surprised mildly, take the half an hour around the FOMC release, look at the change in the 10 year US yield and ask if I run a regression of this on the policy surprise, what is my R squared, okay? Do it on the three month yield, your R squared is essentially like one, right? On the 10 year yield, it's about 8%, which makes you ask, what the hell is the other 92%, right? So it turns out the other 92% is also the FOMC action, but it's there because the FOMC doesn't just come out and say 25 basis points. It comes out and says 25 basis points because we foresee headwinds, blah, blah, whatever, bad things, good things, but there's a lot other than the policy action announcement in there. So in work with Brian Sack and Eric Swanson, we said we could measure this, we can quantify this. We can think of monetary policy as a two-dimensional animal where the central bank does something and says something and saying something actually has a lot of effect on longer term yields. Now this is a, we used to call this target and path, path is now what is called forward guidance, but that language hadn't been invented yet. This is a very particular case of central bank information where the central bank informs the public of its own future actions that it perceives itself following. This is a very natural way of central bank having superior information. It's not about the state of the economy. It's about the state of monetary policy, okay? And that was, I guess, the obvious starting point to think of, okay, what are the things that the central bank is signaling? But then you could ask, okay, why is the central bank signaling a, say, tighter policy path than I expected? If the answer is, you know, we want to get to our inflation target faster, fine. This is a good old-style monetary policy shock. But if the answer is, well, because we think that, demand is going to increase much faster than you think, and therefore we're following our own rule exactly, then there is no monetary policy shock in the VAR sense. This is purely a surprise to you because I hadn't foreseen demand increasing as fast, okay? Good. Now, mechanically, these two litatures were two separate litatures that converged with the invention of proxy VARs. You can think of these as, you know, instrumental variable VARs, which would not be exactly correct, but not too wrong either, okay? And, you know, Stock and Watson round and others, they have been using these things, and those two things, for the purpose of monetary policy, came together with the celebrated paper of Mark and Peter, right? They were the first to properly use these surprises in a VAR to inform the policy shocks. Nice. And that paper also has a section that was reminded to me by Peter yesterday on Fed information, where they say, well, okay, we find these effects, but, you know, let's think whether we're finding these effects off of the surprises in a identified VAR, because the Fed signals something that the markers don't know, so they control for the difference between the consensus forecast and the consensus survey, the private marker survey, and the green book, right? Nice. Now, this paper, then, in the history of thought, essentially picks up where Gertler and Karadil leaves off and says, let's look into this information problem more deeply. And the way they're gonna do it is to say, you know, we're gonna look at stock bond correlations because we understand that positive information pulls stocks in one way and negative information pulls it in another way. Negative information is central bank just raised interest rates for no reason, right? Positive information is central bank raised interest rates and said, because, you know, we are entering this period of wonderfully fast growth, okay? Good. So I'm not gonna repeat because the paper is well written and one can immediately assume that the presentation will follow the same pattern, which it did, but they look at the US and your area, find that the separation of pure surprises versus information matters, right? And then there's a model which I don't think is the most significant contribution of this paper, but it's there and it's useful, and it is actually useful to think separately about how much of the effects we see are due to stickiness and how much are due to financial frictions. Now, it is a good paper, it's worth reading. Getting to whining, all right, yeah, I have things to whine about. One is it's actually worth thinking about what is the relationship between monetary policy and stock prices? This paper takes a very strong stance on fundamentals-based stock pricing where it's essentially something like the Gordon growth formula, right? You'll look at dividends, you're informed that dividends are going to go up when the central bank raises interest rates and therefore stock prices go up. Jammu was complaining about this yesterday, so let me follow in his footsteps, but he would complain more because particularly in this case, right, Jordi Gadi has been on a roll on this kind of papers now where if you believe that there's a rational bubble, rational bubbles grow at the rate of interest, so interest rates go up and I tell you that the world will be great, okay? Well, it's gonna impact the stock prices in the opposite direction that is needed for the identification of this paper. So that's something that, you know, I don't know what you can do about this, but it's worth more than a footnote to say, you know, in the literature, actually there is a lot of work that says this identification will fail and it's actually a lower bound on the information effects that you are finding, right? If you think of non-rational bubbles, then, you know, whatever, all right. So, but that assumption, I think, should be a clearly articulated assumption. The other thing is it's actually really important to think about what one does when one backs out surprises, shocks, whatnot in market-based mechanisms. For example, my work with Brian and Eric and, you know, other people too, when you measure the monetary policy surprise from market-based indicators, right, you're saying, okay, the surprises for these people, the market-based changes show me their behavior change and I therefore measure their surprise, fine. On the other hand, when you do the same thing for central banking formation revelation, you can think of doing this in two ways. Let me give you the example of, again, the central bank's path. It might be that the central bank had in mind, you know, we're not gonna change the path much, but they bungled the words and I perceived a huge change or they didn't bungle the words. It's just that I didn't know how to read it, okay? You're gonna see a huge path factor because the markets perceive that, ooh, the language has changed, okay? They didn't mean to. You could think of, for example, using text-based methods to back out what the central bank says. Assume those things work exactly right, okay? They actually capture the meaning that the central bank is trying to convey, right? The market-based method and that may or may not overlap. They measure two different things. One is what the central bank is saying, the other one is what the markets are hearing. Now here, that becomes really important. There's work by Giovanni Ricco and Silvia Miranda-Grippino that says, you know, let's take these market-based surprises that are also used in this paper and regress them on the green book and the past green books and the changes in the green books which are not observed by the markets at the time of that release, right? We're gonna think of the fitted part of that as the information revelation and the residual as the pure surprise, okay? Which is not too different from what Mark and Petter did in their paper in spirit. But it is very different from what is done in this paper where you don't condition on the actual information of the central bank at all. What you're measuring is the information perceived by the markets. If you were to run an event study regression afterwards, this would be fine. If you were to ask, you know, what does the perception of this information do to asset prices? This is fine. But if you're gonna ask, what does it do to the real economy, then you better be correctly doing that decomposition. It better be the case that the signal received is the same thing as the information signaled by the central bank. This method does not guarantee that, right? So there is nothing in this method that forces the stock bond correlation based decomposition to be also correlated with the information that the central bank has or is trying to convey, okay? Thus, what is happening here is if the market is misperceiving the central bank information, right? And you're using this to inform VIR shocks, these are not going to be good instruments. You're going to be using a very noisy instrument to proxy for your VIR or proxy in your VIR, all right? The other thing here is the following. The paper has a very nice bit that says, you know, the Fed wasn't always issuing these statements. So when we look at the stock bond correlations in the non-statement period and the statement period, they're actually kind of different and that difference is informative, okay? I agree, but then, you know, the gold standard of that is in this building because you always have the release preceding the statement, right? So for the purpose of the proxy VIR, the two event windows for the euro area, which are the press release and the press conference, they're essentially combined together, which is the right thing to do for the purpose of this paper. But to think about how this is working out in action, it's actually very useful to separate these. And a great new paper that will hopefully be written by Carlo Altavilla, Roberto Motto, and others that I know because I'm one of the others, actually separates this and, you know, looks at these in detail. So using that data set, and one of the things that we do there is, you know, we're building this euro area intraday event study database that's going to be a standard resource and be updated, right? So when you look in the press release window, this is what your stock bond correlations look like, right? And it doesn't actually look all that much like the US. They are not like, you know, 50-50 all over the place. It actually looks like a good old fashioned monetary policy surprise with some noise, right? It's really, and I think good for this paper, and I would have shown this, right? It's really in the press conference window that these things are all over the place, right? When the president begins to talk, then the stock market begins to take a life of its own. That is not all that correlated to the monetary policy indicator that they use, and I replicated here. All right. This is minor, but I'm gonna say this. There's a confounding factor that we were careful to think of, and you should too, which is smack in the middle of that press conference, the US initial claims is released. And the initial claims affects US stock prices, but also the European stock prices and the OIS here. So it's gonna affect your correlations and that's something worth controlling for. The R-squares are low, but the R-squares in these information regressions are also low. So we understand that, apparently, those low R-squares does something large in the VARs. It's worth controlling for. All right. I'm gonna leave with this. What the hell is the central bank information? What is it that a lot of people who are central bankers, I'm gonna ask Frank, what is it that you know that other people don't? This is actually really important. It doesn't have to be answered in this paper, but it is important that we think about whether the central banks know something about the future, know something about the now cast, know something about their own preferences, know something about the steady state, but what is it? And this is because this rumor and rumor that gets mentioned all the time, it's actually that result isn't there in the Great Moderation period. Central banks do not have an informational advantage in inflation forecasting during the Great Moderation period. And the Great Moderation period is the sample of this paper. In fact, the Roman and Romer result is almost entirely due to the Volcker disinflation where he comes in and says, we're gonna disinflate whatever the bloody cost and then tells the staff forecast inflation and they say it's gonna fall and yeah. So it's not clear at all that the central banks does have any informational advantage in forecasting, maybe some in now casting, but these are worth thinking about. And the other thing here is, I'm gonna skip this, that makes it important not to refer this as a separate tool. It's not something that the central bank can exploit. The paper uses this word and I actually would shy away from it as much as one can. Because if it's a tool, then it becomes my question whether I want to correctly inform you, whether I want to give you the positive information or not, but then you're not stupid, you're gonna figure out that if I'm not saying much, probably things are worse than I expected and that general equilibrium is not a pretty general equilibrium. So, but it's useful to think of whether this is a tool or not, whether it can be avoided, whether we want to use this, whatever, not this paper. On the other hand, something not in the paper but in Marek's discussion or presentation was it's really important that we think of whether the link is causal. Is it that, because the VAR is a causal mechanism if it's an identified VAR, right? So the early part of the paper reads as if, because the central bank comes out and signals that bad things are going to come up, look bad things are happening. That's probably not the case. That's not what the model is suggesting, right? But then we have to think of what is the case. All right. I'll let you read the conclusions and thank Frank for his patience. Thank you very much. Thank me. Thank you very much, Refit. So we have 11 minutes for discussion and answer. I suggest we get some questions and then I'll give the floor back to Marek. Annette and then Sevnan. I agree with Refit's comment that I don't think the central banks are actually that better informed than Deutsche Bank or Goldman Sachs. I think the main problem with this whole literature about the Fed information effect is that the Fed funds futures are a completely messed up measurement of monetary policy because if monetary policy is effective then suppose you had a case where there was only a good and a bad state and the central bank said, we will do whatever it takes to make sure the bad state doesn't happen. Then well in a good state you're gonna have by a tailor rule the interest rate is gonna be higher. And so you could see that a very accommodating monetary policy that basically made the bad state never happen would show up as an increase in the federal funds rate. And that's why when you look at the quadrants you can get points where the stock market goes up, the futures go up. It's precisely because monetary policy works not because the central banks are better informed. Very nice paper and very nice discussion. I'm an outsider to this literature but I find it very interesting that the European central bank authors focusing on the US and why not the Europe where exactly as Refet said these two different windows is the gold standard in this building. So you have the statement in I guess this council meeting at 1.45 and then there's the press conference later. And in fact, we just heard from Refet that he's working on it. And I recently saw a paper by John Rogers which is ironic because they're all fed co-authors. John Rogers and Andy Kane and there's a third fed co-author and they actually exactly do this. They do exactly what you're trying to do and what Refet just mentioned. They are doing try to understand the surprise and the information content of the ECB using these two windows. And one interesting thing they find is this goes to the last comment Refet but what the hell the central bank knows they do find this perverse wrong direction effect in the information window. The GDP goes the wrong way like Nakamura Steinsen but they also find this using the OMT and then the European crisis that preserve the euro effect. So with an expansion monetary policy you have the appreciation but that happens not through the statement but not through the original statement but through the communication windows later. So maybe you can also look at this ECB policies and then compare to the US. Thank you. Third, anybody else? So let me give floor back to Tomarek and then we'll see. Thank you, thank you Refet for your discussion. It was very thought provoking as usual. So yeah, so you'll give a few reasons why our identification shouldn't work on the rational babbles and the volatility in the stock market. And that's why we also started this project full of doubts but the answer is in the end the simple restriction it does separate a contractional shock from an expatriate shocks. So yeah, so it's surely not, I mean it's not the last word no one could and should refine this further but and we have made a few attempts and I'm sure other papers will also do it but the point that we just want to move the message that we want to leave you with is that this one simple restriction is enough to separate an expansionary from a contractional policy shock whatever noise maybe still remaining in both of them. So we're very sympathetic to this paper by Miranda Agrippino Enrico and others who are and actually also Peter and Mark have done a similar exercise in their paper to extract the fed information from the differences between the fed forecast and the private forecast and this is, we think of this as complementary work and what gives you a different measurement of different approximations of the same effects. And then you mentioned the initial jobless claims and we do control for them and then they change very little but we do it just to be sure. So now we have Shabnam thanks for the suggestion and then on the comment of Annette. So the way I think you have in mind such an idealistic situation in which the central bank perfectly neutralizes shocks and that's why you don't even see the effect. So we think of a central bank that moves rates gradually and it's common to find that when people estimate policy rules it's common to find interest rates moving so that the central bank moves interest rates in the right direction but not enough to completely offset the effects. So that's why you still see a correlation in practice. Thank you. So this is a great paper and a really excellent discussion. I would like to ask you about your reaction to Otmar Ising's speech at this S&B last week who also stressed on communication central banking who also stressed the importance of communication but importantly sees it as part and parcel of the central bank policy policy strategy. So this brings me to the question of challenging your assertion at the author that there could be an independent communication policy. So my challenge there is to say, you're trying to say there is an independent information shock and like Annette I would say central banks don't really know much more about the current data but a way to reconcile is what happens in the press conference is actually an updating of the reaction function. So the whole communication purpose is to improve the understanding of the public that is to reduce the friction because it's intended signal and the received signal as your as a discussant was saying and that's the whole point. That's the point I've made in a paper 20 years ago myself but with a nice quote from Wittgenstein as a philosopher so I'd like to go back to that modeling. And then the second issue, so one is it's really about reducing information friction and how to measure that. The second point is the shin argument about beauty contest and echo chamber. So there could be a detachment and that's a bubble kind of thing that we go away from fundamentals if markets react to central bank communication and in turn central banks start reacting to market reaction to their communication and I'd like to have your reaction to that I think you were very careful in your conclusion to say you cannot rule that out but I think that's a key issue that you need to put in the modeling as well. So maybe I'm breaking a bit the typical link between the surprise and the systematic response. I think the key thing is actually getting the systematic response across. Okay, any last questions? Maybe if I can just add one myself. I guess the question is where do we end in trying to decompose these very sort of, I mean short and the small surprises? What I mean is now you have a monetary policy shock and what you call an information shock but of course the information shock that could be decomposed into information about supply shock or information about the demand shock and of course the effects of the supply shock will have very different effects on the stock market than the demand shock. And so we can add of course other information, we can add inflation swaps to try to decompose supply versus demand and probably somebody will do that. You should read the paper, it's already there. Oh, sorry, sorry. Okay, so that's the answer. So where is the end? I don't know where is the end. So, yeah, and so also, but this is where the literature is actually heading. Now, so there is the recent paper by Barbara Rossi and Inua who actually identify the effects of policy but recognizing that there are no two shocks are alike. All shocks have different nature or different differently move, shorter pieces of the yield curve, the longer pieces of the yield curve. So they go much more granular than we do. So I think we're not yet at the frontier. And coming back to Bernard's question, so maybe the goal of communication is to update the public understanding on the reaction function. However, so information about the reaction function that's all captured in what we lump together as mantra policy shock, right? Because updates of the reaction function will induce this negative correlation between the interest rates and stock prices. So for our mantra policy shock captures current policy, forward guidance and updates about the reaction function. Okay, I think we're right on time. Time for lunch. Same place.