 So good afternoon everybody and welcome to this first session of the conference on the relation between monetary policy decision making and financial markets. So I'm a man. I mean, you will so I'm the director general for market operations at the city. And, and this first paper that Michael power will present is really looking at an alternative explanation for the fed information effect. This is a paper that, you know, the Liberty adopts, I think a mid busting approach and gets into the crux of a question that we as financial market professionals ourselves routinely with is, you know, what is exactly the information content of central bank decisions. Yeah. And does the central bank know more than the market, or is it just reacting differently than expected to the same set of information. And in other words, should we read into a fed decision or any other central bank more than it says, and I'm delighted to have with me, Michael power, one of the authors to present the paper. So Michael, you are a professor of financial economics at the University of Hamburg since 2020, but you started your career after your PhD from the University of California San Diego as a central banker. Yeah. And you were senior economist and research advisor at the Federal Reserve Bank of San Francisco for about 10 years. Your research focuses on the interactions of financial markets with the macroeconomy and monetary policy and your work has been published in leading economic journals, including the American economic review, the review of financial studies and the journal of business and economic statistics. Michael, the floor is yours for presenting your paper. You have 30 minutes and this would be followed by 10 minutes of Q and a with the audience on the chat. Yeah, thank you so much for the kind introduction. Toughly to follow here with such a great presentation by Ken Rogoff. I'm going to go ahead and share my slides here. I think you all should be seeing that and hopefully you can hear me all right. Okay. This is work with Eric Swanson. We both used to be at the Federal Reserve Bank of San Francisco for a number of years and the motivation here is really coming from, you know, our experience at a central bank where we were kind of doubtful of this story of central bank information effects that there is so much superior private information at the central bank about the economy. Certainly, we as forecasters were looking at the private sector just as as much as maybe the private sector was looking to the Fed. And so we were looking at the forecast and thought, you know, there's highly resourced and experienced and experts looking at the same data that we are looking at. And so we were skeptical of the story that the Fed is superior in its knowledge and forecast of the economy. And so we dug into the data to kind of see whether this evidence on these information effects is really as solid as it seemed. And so we found that it is actually not. So we came up with an alternative explanation for this evidence. Okay, but it's all about these monetary policy surprises that are widely used in empirical macroeconomics. So those are essentially rate changes around FOMC announcements or ECB announcements. Think about, you know, money market futures rates, take a handful of them with maturities of several months to several quarters, and the 30 minute changes around the announcements. And then people usually take maybe the first principal component of these rate changes to summarize what's going on. Now, the usual assumption is that these are predetermined and unpredictable. So then you can use them because they're econometrically exogenous to estimate the causal effects of monetary policy. For example, on financial markets, and a number of papers have done that. And more recently, people have even used this to estimate the impact on the macroeconomy using things like structural VARs with external instruments and so forth. Okay, so let me just make sure my volume is high enough here. Okay. All right, so now there is this really puzzling recent evidence that the response of macroeconomic survey forecasts seems to go in the wrong direction. So for example, a tightening surprise, so an increase in these futures rates around an announcement would, according to standard New Keynesian theory, lead to a negative response of output, employment, and inflation. And so you'd think that survey forecasters would revise down their outlook for these variables in response to a tightening surprise. But what a number of papers have actually shown is that the survey forecasts respond positively to the monetary policy surprises. Maybe most prominently a paper by Nakamura and Steinsen in the QJE has shown that real GDP growth forecasts respond positively to these monetary policy surprises. There is some related evidence about the stock market that I may or may not have time to talk about. But an explanation for the evidence, for the puzzling evidence from surveys, as well as for the evidence on the stock market is a fed information effect. Okay, so what does that mean? That really means that kind of independent of the actual policy action and its impact on the economy, the central bank conveys private information about the economy and directly affects beliefs about the economic outlook with its announcement. Okay, so as an example, if you have a tightening surprise, so interest rates jumping up as a response of the announcement of the central bank, then this might be seen as good news for the economy because it could be seen as signaling a more optimistic outlook of the Fed, the central bank. And so forecasters and investors might then revise up their outlook, taking that signal on board from the central bank. Now, these information effects are really strong than that has really dramatic implications for both empirical macro, but also the practice of monetary policy. I mean, just to sketch them, you cannot use policy surprises, these rate changes anymore to estimate the causal effect of monetary policy in the economy because you have these confounding information effects. And they would be really hard to account for because by their nature they are unobserved. And for the practice of monetary policy, and you certainly see concerns like this in policy discussions, for example, in the minutes of the F1C meetings. In practice, it might mean that surprises may have counterproductive effects. And so policy makers might be less inclined to surprise financial markets than in the absence of such information effects. All right, so we propose an alternative explanation in our paper of this evidence based on what we term a Fed response to news channel. Now, what this means is that, well, of course, survey forecasts and monetary policy actions both respond to macroeconomic data and macroeconomic news. But if the monetary policy surprises are also systematically related to the state of the economy, then you have a problem in these regressions and the basic underlying assumption of a lot of empirical work using these surprises is violated. You basically have an omitted variable and we'll show that that matters a lot. So our evidence is, first of all, going directly back to the survey regressions and showing that economic news is an omitted variable and accounting for it changes the results dramatically. We also did our own survey of professional forecasters. We went to the blue chip panel and asked these forecasters how they respond to monetary policy and our results suggest that they respond in a very conventional way and that there's almost no evidence at all for the possibility of information effects from the forecasters themselves. We revisited some of the financial market evidence and we don't see any convincing evidence there that there's systematic information effects. And we also revisited some of these forecast accuracy comparisons is the Fed really a better forecaster than the private sector. It does not seem like it. So overall, we find no evidence for information effects in FMC announcements. And we have a story for why monetary policy surprises may be systematically related to macro news that is based on incomplete information about the monetary policy rule. But that is kind of not crucial for the main points I'm trying to make here. Okay, so let's look at this survey evidence. Okay, so this is the regression that Nakamura and Steinsen ran revisions in forecasts for GDP growth are regressed on their monetary policy surprise. They just use the first principle component of changes in these futures rates around the announcement. Okay, so under the assumption that this policy surprises exogenous, this would estimate the causal impact on the forecasts and standard theory implies that the impact should be negative. And and puzzlingly they estimate positively and significantly positive as coefficients. Okay, so we're going to revisit this here in our sample. We updated the data. We looked at different surprises, different macro variables. And overall, we also find this puzzling sign, GDP growth and inflation respond positively to a tightening surprise. In the upper panel, we use this Nakamura Steinsen surprise. So the first principle component of futures rate changes around the announcement in the bottom panel. We used the more commonly used target and path factors that were proposed by Gerkinak second Swanson in 2005. And for the unemployment rate is the other way around. Okay, so there is something there, but it is really not very strong. There's some warning flags here. This these results are often insignificant. They change a lot over different sub samples and their statistical relationship is very weak. The R squares are sometimes round to zero or generally between, you know, one and 5%. And so this is very noisy. But still overall there seems to be something there that we need to look at a little bit more closely. By the way, we of course replicated Nakamura Steinsen's results and looked at different samples and investigated how this changed with the sample choice. Okay, so our story is that there are economic news that are affecting both the FOMC action and the survey forecasters, right? So you have a good employment report. The surveys are respondents are going to revise up their outlook in this example. And the Fed is of course generally going to respond by increasing the policy rate on the margin if there are good news. Now if it responds systematically more than expected. So if there is a systematic hawkish monetary policy surprise when there is good economic news, then you have a problem with these survey regressions. Because then you have an omitted variable bias. Okay, so the usual regression is the I showed before, but there's also these economic news and omitted variable. Now you have a bias from your econometrics 101 class. If the news are correlated both with the dependent variables of the survey forecasts, which of course they are, right? There's no question about it. We have extensive evidence of this also in this paper, but it's really not surprising. And I'm going to skip that evidence here in the interest of time that economic news, economic data affect the survey forecast. So beta is of course not zero. But the more puzzling issue here is you also need the news, the data, the macro data to be correlated with the policy surprise systematically. The usual assumption is that this policy surprise is not correlated with data timed before the announcement. And here this, I'm a little sloppy with the notation here, but these news are all measured before the announcement, before announcement T. So known at the time of the announcement and the policy surprise is the change in rates around announcement T. And so we find that gamma is also not zero. So let me show you that. And then I'll include these news into the baseline regression and show you that there's a big bias in theta. Okay, so this is the data in Nakamura Science and then already kind of tells the story. There are big positive, you know, revisions to the outlook on the upper right that were also that also saw big hawkish policy surprises. So they're systematically hawkish surprises when the economy is doing very well in the late 90s, 2003, 2004. And on the other side of this, these most influential observations in the lower left were times when the economy was doing very poorly. So of course, the outlook kept being revised down and the forecasters, that's not surprising. But what is surprising is that there were systematically large dovish or easing surprises. Now, there seems to be this omitted variable here, the state of the economy, whether there's good or bad news about the economy that explains this positive correlation and not a causal effect from the announcement surprise to the macro forecasts. Remember, that is the story that several papers are telling. There's a causation from the monetary policy announcement to the macro forecast. And we're saying, look at this picture. There's an omitted variable that affects both of these. Okay, so this is the one I'm going to skip. Right. And then here is the slide with, I apologize, a little too many numbers, where we regress the policy surprise on a bunch of macroeconomic news and financial indicators. All right, so three different regressions for three different kinds of policy surprises, the target and path factors, and the Nakamura-Sanson surprise. And without going through each of these individual predictors, I'm just going to summarize this evidence as saying these policy surprises are predictable with information before the announcement. The predictability is very strong and some other samples that goes up to 40% are squared. This is in line with some evidence that the next presenter, Anna Seeslack, published in a 2018 RFS paper. We kind of revisit this, look at a few more predictors, find a little bit stronger evidence, but very much in line with Anna's findings. Now, but what we conclude here is that there is a problem with the basic survey regressions. We have to control for these macro news. We took these financial indicators, by the way, because they really nicely summarize all kinds of news that we couldn't possibly all put into a regression. The first one, for example, is the change in the S&P 500 log. So it's a return over roughly the quarter leading up to the day before the FOMC announcement. So that is a positive return. That is usually a time when there's a lot of good news. And that then systematically and strongly predicts a hawkish policy surprise, which is puzzling from, you know, kind of like a standard theory perspective. Okay, so this is the key result on the survey regressions. The top panel just reiterates the basic results of the univariate regressions or, well, the regressions without any controls. So you have these kind of puzzling coefficients that sometimes are significant, not a very strong relationship though. But if you include these controls, then essentially all of the coefficients flip signs, the magnitudes are equally large, if not larger than before. And often, strongly significant. The R-squared, of course, is very large because of the controls that we include. So you estimate these also much more precisely the effects of monetary policy surprises. So here, the upshot is if you control for the state of the economy, for the economic news, then monetary policy surprises have a totally conventional impact on macroeconomic forecasts, okay? So we think that this is a better estimate of the causal effect of monetary policy on macro forecasts. And that it suggests that information effects, while we can't rule them out completely, are not strong enough to make these coefficients zero or even have the puzzling opposite sign than conventional theory would predict. And the coefficients here, I'm not going to go through the units and the sizes and the quantities, but they're generally in line with what you'd find from standard textbook monetary economics models and estimates of the effects of monetary policy on the macroeconomic. Now, why are monetary policy surprises predictable? We really think it's because of learning about the policy rule. We don't think it's a risk premium story. Risk premiums are really too small in these near term futures. It's not crucial for our story, but we think that it's incomplete information about the policy rule. There's some supportive evidence and I have some evidence in her paper, a recent working paper by Mike Schmeling and co-authors also suggests that the market underestimated the Fed's responsiveness to macro data, systematically underestimated, which is exactly what we need to explain our results, okay? But I'm going to leave it at this, you know, this does have really strong implications though for applied work. You cannot use monetary policy surprises as an exogenous instrument if they're correlated with the state of the economy. So you need to account for that, for example, by taking either controlling for the news as we're doing here, or by projecting out the news and just taking the residual of the monetary policy surprise. So this is important for this literature. Okay, so I want to say a few words about our own little survey. The nice thing about financial data is that you can really isolate the effects of the FOMC. That's why there's a lot of papers using event studies, Kutner, Gokainaxx-Wanson, Bernanke and Kutner many more, because you can be pretty sure that you have a causal impact there. But surveys are kind of annoying. They are monthly, quarterly. We wish we had daily surveys. Our solution to this problem of low frequency surveys is to ask the forecasters directly about how the FOMC announcements affect their forecasts. We tracked down the chief economists of the entire blue chip panel and sent them our short questionnaire. How do you revise your macro forecast and response to these different components of the FOMC's policy action? We also asked them about how they revise in response to the SEP forecasts. I didn't include that here, but it's just shocking. They never respond to the SEP forecast, which is really a problem for the information effect story, because that says the Fed has better forecasts or better knowledge, or at least somehow useful knowledge, somehow not a perfectly correlated signal. So people should take it on board, but they don't. At least that's what they tell us. Okay, but back to the policy action. What if there's a hawker surprise and either the target rate or the statement or the dot plot? Forecasters either don't revise their forecast or they revise it downward, which is the standard direction for a hawker surprise. Everything here is mirror image for a double surprise and everything is parallel for the other macro variables. A handful said that the direction depends on other factors. But that would be the only forecasters that you might count in the camp of sometimes potentially some information effects. Even with this generous counting, you'd have 31 to 5 against information effects from our survey. Now remember, this is exactly the survey forecasters that are delivering the data that are used in the regressions that are the information effect regressions, if you will. So we think this really undermines those kinds of regressions. Okay, so I think I may have two or three minutes to talk about financial market evidence. So this is a table with the 10 most influential observations in the Nakamura Steinsen regressions. So these are the ones in the upper right and the lower left of my scatter plot. These really drive, like, sorted by their effect on the T-sit to see if you include them versus exclude them. Right, so of course, you know, the policy surprise in the third column and the GEP forecast revision in the fourth column have the same sign. These are the most influential observations. Now, what did the stock market do around these observations? It essentially always went in the standard direction of a tightening surprise causing lower stock prices and vice versa. Okay, there's only one exception here in March 2001. And similarly for exchange rates, exchange rates also went kind of the standard textbook direction of a tightening surprise appreciating the dollar. Right, so this kind of high frequency evidence suggests that these announcements did not actually contain strong information effects. If you look at the tight window, in that instead what caused this positive correlation between this policy surprise and the macro forecast was the state of the economy. Summarized here in the last column with this Brave Butters Kelly macro indicator from the Chicago Fed where negative value shows the economies in that state and vice, is in that, is in that shape and vice versa. Okay, but that was basically our story for explaining the survey response. Okay. Now, let me just say that when we regress this, you know, we put it this way. Information effect story does not say the stock market response necessarily has to have the opposite sign. It just says because of how stock markets work and discount rate news and cash flow news. It's just, it just says there should be a weaker response if information effects are there. Okay, so maybe they are just weaker, although still of the conventional sign. What we did here is we kind of split the sample into the most influential observations. So presumably, they would have big information effects and show a weaker stock market response. But that is not the case. It's at least as strong, not stronger than for the other observations in the sample. Okay. All right, so let me conclude. Let me just conclude here. I think that overall we see quite extensive evidence against information effects in FOMC announcements. Okay, so we haven't looked at ECB announcements. We also haven't looked at other policymaker speeches like chair, chairman speeches, chairwoman speeches. I'm trying to find some more information effects there. So maybe there's something there, but certainly around FOMC announcements, there doesn't seem to be any convincing evidence for information effects. The standard survey regressions have a big problem. They have an admitted variable bias. And if you count for these admitted variables, then you resolve the puzzle that has been documented in several papers. And additional evidence, I talked about our survey, I talked about financial markets, and I just want to emphasize again that the Fed is not a better forecaster than the professional forecasters. If you look at the blue chip forecasts, for example, and so that is also undermining the story of a better informed Fed. Now, we find that the policy surprises are exposed. Now, this is not a real-time out-of-sample forecasting exercise. It's just a correlation in the sample. Exposed, they're correlated with economic news. And this correlation is consistent with the market underestimating the responsiveness of the Fed. For macroeconomics, this means, for empirical work in macroeconomics, this means that monetary policy surprises should not be treated as exogenous. Ideally, you would want to project out macro news. Eric Swanson and I are working on a follow-up paper that we hope to present at the NBR macro annual next year, where we do this analysis. So revisiting local projections or structural VR estimates of the effects of monetary policy, it's important to account for this channel that we document here in this paper. And I talked about the implications for monetary policy communication. It's not a concern from our evidence, at least, that surprises would have counterproductive effects because of revealing information effects. So we think that's a first-order issue for monetary policy communication. That is all I have at this point. Thank you very much.