Hello everyone. I'm Ronnie Sadka. Very happy to be here. I want to talk to you a little bit about central bank monetary tone. The team and I - the MKT MediaStats team - has been working in the last several years with the GM - global markets - team in stage 3 to understand the tone about different financial assets. What we'd like to talk to you about today is particularly about central banks. Now, what's the problem that we're trying to tackle? The problem with central bank communication is that recently they've become more fake or uninformative. Why? Because central bank members understand that they are being looked at very carefully and the minutes are being looked at very carefully. So when people and machines are looking at the words and trying to understand what they mean, that feedback effect in anticipation of NLP algorithms looking at their communication, they choose their words very carefully. Therefore these minutes become more opaque. The other problem of course is that when you're interested in central bank communication, it only happens several times a year. Our solution to this problem is to look at media coverage. There's a lot of media coverage that talks about central bank communication, monetary policy, and it happens throughout the year. By looking at media coverage - so we look at the media from our reservoirs of information, hundreds of thousands articles - we're able to tease out globally what the world is saying about how they're interpreting central bank communication and the actions that they take. In addition, because we're looking at media coverage, we have indicators that we produce every day. Every day we have a view of central bank monetary policy. So what is it that we do? What we're going to do today: we're going to look at, how do we measure central bank communication? We talked about some of the properties and applications of the tones that we're going to measure. We're going to extend the measurement of central bank communication. We're going to extend that to the individual members so we're actually going to look at the sentiment or the tone per member. Then we're going to qualify our results by looking at three things. First, we're going to look at the average tone, so overall what is the Fed or any central bank saying about rates? The second is, we're going to look at individual members. We're going to show whether the chairman of the Fed has additional information, or what the media is perceived of their view in addition to the overall bank. The third is, given that we have information about individual members, we're going to look at whether or not these disagree or agree. The amount of agreement, we're going to measure that as a standard deviation of their views. We're going to show that that also has information about yields. So let's jump in. How do we measure central bank communication? We look at our reservoir of articles and we classify articles as to whether they're hawkish or dovish. What does it mean, hawkish? Well, we look at words like interest rate hikes. We look at tightening, whereas dovish, we're looking at more easing, rate drops. There is a bag of words for each of these, hawkish and dovish. Each article is classified whether it's hawkish dovish. Then the way we construct the indicator is by looking at the difference between the number of articles that are hawkish minus dovish, divided by the sum of all articles that are either hawkish or dovish. At the end of the day, it's an index. It's a score that if we see the values of between minus one and one, one is very hawkish, negative one is dovish, and then all in the middle. You can see on the graph on the left side the sixteen central banks that we cover. It's the G10 plus some emerging market central banks. What's nice about it, you can see there's a significant cross-sectional variation. All of these numbers here are a snapshot of October 2020. In this case, the most hawkish was the Bank of Brazil. The most dovish is the Central Bank of Turkey. The nice thing is, actually once we saw this in our media monitors, in our indicators, after that the central banks - actually Brazil - increased rates and Turkey reduced rates. So that was interesting. Overall the data, the time period that we have, that we produce is from January 2015 through October 2021. Again everything is point-of-time data. We have a healthy amount of articles. For the Fed communication we have a 240,000 articles during this almost seven-year time window. ECB is 94,000, Bank of England, 56,000. There is a sufficient amount of information to calculate meaningful indicators. Let's take a look at what these indicators tell us about rates. In this graph what we see is the central bank tone for the Federal Reserve. We are comparing that to the three-month Treasury Yield. What you see here, here's what's interesting; that when you look at central bank media indicator, what you see is that it has been earlier in the sample more hawkish and it became more dovish from the end of 2018. Then we see towards the end of 2019, again there is an increase in hawkishness. Then when the pandemic hit, the lowest hawkish score is around March, April of 2020. Since then it's been increasingly more towards hawkish. Just from this datapoint, when you look at the three-month Treasury Yields, you see at the end of 2018 for example when the media suggested that the Fed is becoming more dovish, rates still went up. Then they started moving down with the lag, so we do think that this indicates that there is some predictability that the media tone - hawkish-dovish indicator that we have - can tell us something about future rates. To do that, to test this we're going to run some regressions. What we're going to do is the following: we're going to take the time series of the hawkish-dovish indicators. This indicator is a seven-day moving average so we're going to look at let's say on a Friday, the last seven calendar days, every day we're going to calculate the average hawkish-dovish of the articles that cover the Fed. We're going to average that over the last seven days. We're going to be using that to predict future rates. We're going to look at the changes in the yields. We do it for several - I'm going to show you later soon - the entire yield curve but we have here a panel for the one-month change, three-month change and six months. We do that for every single week post-the week where we're measuring the hawkish-dovishness. The way we run this regression: every column is a single time series regression of future rates on the hawkish-dovish indicator. We do a bunch of controls. For example, we control for the level, the rates and other variables. The key message here is the following: that when you regress yields on past hawkish-dovish indicator, you do see a positive and significant coefficient t-stat around three. Now, we also differentiate between Fed and FOMC meeting weeks versus non-FOMC meeting weeks. We do that with a dummy variable. When you look at the box there on the top, that indicates the hawkish-dovishness has a positive predictability on future rates during weeks where there is no meeting. When you look at the weeks with the meeting there is actually not much differential effect. You can see on the bottom there, certainly in the six months, the one week for change in six-month rate, it actually completely offsets. You see that there's a plus three and a minus three. That almost offsets even though it's not statistically significant, that difference. To better explain these numbers, we're just going to graph. What we do here is we have the yield curve and you have the short-run rates, long-run rates, all the way to ten years. Each line here is the change in rates on the three-month… Sorry, not three month; we have the entire yield curve, but the three-week-ahead change in rates post-the week where we're using the indicator. The blue line shows you the change in rates and basis points over the next three weeks for weeks that do not have any FOMC meeting. The line below is the one with the FOMC meeting. The conclusion is that the effect is more pronounced throughout the yield curve for non-FOMC meeting weeks. To us, it makes sense because these are the weeks when there is no central bank communication. What seems to be happening here is that the market underreacts to the information in the media. Therefore the media has a positive predictability on future yields, so that's about the central bank. Let's now look at the individual members. Let's extend the measures to the individual members. We're going to look at the chairperson, Jerome Powell. We're going to look at other FOMC voting members. What you see here is a snapshot as of October 26. What you see here, again a very nice cross-sectional variation across the different members. Tom Barkin there, the most hawkish here, and Kashkari is the most dovish. Powell is somewhere in the middle. Again because the media coverage is more sparse for individual members, we actually look at a thirty-day moving average of hawkish-dovishness measured per person. Now let's look at, on the next slide what we're going to do is we're going to compare overall Fed hawkish-dovishness to, for example, just the chairman. On the left what you see is, in blue you see the overall Fed communication, Fed hawkish-dovishness indicator. In red what you see is Powell's indicator. What you see immediately is actually they're correlated. They move together but they are not identical. On the right you see the differences between tone of the chairman of the Fed and the overall tone of the Fed. There are differences. I think the natural questions is: well, are these differences informative? What we're going to do is, we're going to run a regression - the same regression we did before - but we will add now in addition to overall Fed tone, we're going to add the tone of the chairman himself. The way we're going to do this is, first we're going to run a regression of chairman tone on Fed, take the residuals, add the residuals in these regressions and also interact these residuals with a dummy for FOMC weeks. At the ned of the day, what you have here is, every column here is a single regression where you're regressing future rates, changes in rates on overall Fed tone. Then interacted with FOMC meeting weeks, then residual chairman tone, and also interacted with FOMC weeks. The conclusion here is that the Fed tone, the chairman tone has an independent positive effect over and beyond that is controlling for overall Fed tone. The point is that the chairman has its own predictive power even over and beyond overall Fed. That's an example where chairman tone is important. What about the other members? Now, the way we decided to show this is first just to show people, what is the standard deviation of tone across members? When you calculate standard deviation, each measures between minus one and one; the standard deviation here is going to be between zero and one. A higher number means that people disagree, and a lower number means that there is more consensus. First we see that there is significant time variation. Now, I think it's interesting that the least or the most consensus, the least sigma, actually have been; we observed that around the first few months of the COVID-19 pandemic. After that, the disagreement started going up. Then actually in the last year we see that there is more agreement because it's shown that sigma is dropping. What we're going to do is we're going to try to understand whether the variability between the tone of the different members tells us anything about rates. The way we're thinking about it, well, if there is so much disagreement then maybe there is more uncertainty? Indeed when you include the sigma - so this is the cross-sectional variation in tone across members - we find that that has a positive predictability for future changes in rates. Potentially what it signifies is that the market understands this as more risk because some people are positive, some are negative; more uncertainty. More uncertainty; yields go up. Now, to test this further we said, well, maybe what we should do is differentiate between FOMC meetings that have dots without dots, because in FOMC meetings with dots, you can see what different members are saying about the Fed fund rates and other variables. When you have dots maybe there is less uncertainty because you can understand more what's going on. What this graph shows you is precisely that: when there are no dots, FOMC meetings when there are no dots, that's when sigma has the most effect. That seems to be when the market just looks at the media. If the media is saying each person is looking at something else then it's thinking about a different tone; that's when there seemed to be the most predictability on future rates - which again is consistent with more risk. Now, if we run a regression with all of these three, we talked about three things; we talked about overall Fed communication, we talked about the chairman communication and then we talked about sigma. This is the amount of agreement or disagreement across the members. When you run a regression of all future rates, yields, changes in future yields on all three and you look at the relative importance of these coefficients, what you find is the following picture: that the overall Fed tone accounts for about fifty per cent of the effect. Then the chairman versus the sigma, each of them is about twenty-five. The sigma is a little bit higher, but the way to think about it, the way I think about it is half of the effect is really the central bank. Then the other half is equally split between the chairman and the variability between the other members. Just to summarise, what did we do here? We tried to go over, how do we calculate the tone of central bank monetary policy? We show that it's important to understanding changes in short-term rates. We also extend this to individual members. We showed they have even an additional independent impact. Then finally we looked at the sigma or the consensus, or lack there of consensus, and we showed that that's associated with increased risk. That's my part. I want to move this to Marvin, who is going to tell us more about what we see right now with these indicators.Thanks, Ronnie. For better, for worse, one of the most important inputs for us as analysts these days is really understanding the tone of monetary policy. What I'm going to try to do is briefly go through the indicators that MediaStats and Ronnie and his team have put together in terms of how it's a valuable alternative data source in helping us ultimately judge the tone of central bank policy. I'll begin our journey, if you will, by looking at a contra-example where what we're looking at here now is the tone for the ECB relative to bund yields. As all of us who have been in the markets for a while know, the ECB was often considered the most dovish of central banks - even before the pandemic. Once the pandemic hit, for the most part their normalisation path was the longest amongst the G10, with the exception of the Bank of Japan, of course. That pretty much was shown within the MediaStats indicator for the better part of last year, as you can see here. That started to change in the beginning of the year when first it became less dovish and then we moved into a slightly hawkish position. Bund yields followed that path. Of course the ECB pushed back. Most recently Madame Lagarde effectively told the market that they had the pricing wrong; that caused a rally in bund yields, as you can see. What didn't happen was that there was no corresponding dovishness within the media. That hawkish view has been held with the media, which to me provides the potential opportunity for a fading of the rally that we saw in bunds over the last couple of weeks, particularly as we get closer to the December meeting of the ECB that will potentially provide a path with regard to what the ECB is going to do on the tapering fund. Moving a little bit closer to home: within the FOMC certainly, as Ronnie discussed, the chair's voice is the voice that rings the loudest amongst the gaggle of many voices out there. As Ronnie also showed, the market pays attention to that. What I found interesting was that, while certainly short yields have been driven by changes in the tone of the chair, recently over the course of the past year it's been the longer end that actually has driven things. What you could see here is that there is a fairly nice fit within how the ten year moved. The way I interpret that is that shorter term, there is not as much impact that the chair is going to have with regard to the start of the hiking process. That's for the most part going to be driven by the data, by inflation, but longer term the path for rates is more important; how shallow or how long are the hiking processes going to be? The other thing that I see here is that the recent sell-off in the ten year has not corresponded with a more hawkish tone coming out of media - which again provides the potential for a fading of the more bearish view that the market has had on the long end. Now, are there other voices? Ronnie talked about the other voters. I've always looked at the group of DC voters as the second-most important block, if you will. They're the second-most important because all of them vote and they generally vote as a block with the chairman. What we could see here - which is the average of all the DC voters as well as the New York Fed - is that they too have moved slightly more hawkish over the course of the year. But they still remain dovish overall and they still remain neutral. When I compare that to the ten-year real yield, that ultimately makes a bit of sense because once again it's where the path of rates is going longer term. This is saying that if anything, real yields in this negative category are going to remain. It's really formed, for me, one of the backstops that I use in thinking that the curve is going to remain relatively flat at a time when we're really thinking about inflation pushing forward yields much higher. Another thing that these indicators provide, at least for me, is a sanity check particularly as we go into the FOMC meetings. We spend a lot of time really thinking about what the outcome is, when in fact very often the pre-communication that occurs within the media correlates well with where yields wind up. So what we have here is just really the average of the hawkish-dovish indicators for the Fed going into the FOMC meeting, relative to the two-year yield as it comes out of the FOMC meeting. It fits pretty well, which ultimately affirms the overall view that the market has. So there is a lot of pre-market pricing before the FOMC meeting, but more importantly is looking for potential changes. Some of that dispersion and some of that contrast that Ronnie talked about in terms of a change in the message going forward. I'll leave you with just some thoughts about the new Fed… I guess we'll see who's running the Fed for the next four years. For the most part, the conversations have come down to Powell versus Brainard. We saw a fairly significant market reaction when Brainard's name became more official a few weeks ago. What we tried to do here was to look at how the media covered both Brainard versus Powell from a hawkish-dovish perspective. Once again a number that's above zero in our low index here means that Powell is more hawkish than Brainard at any given point in time. A number below zero shows that Brainard is relatively more hawkish. They could both be dovish at that point, but Brainard is sounding a bit more hawkish than Powell. Certainly the indicator's above one, so the view that Brainard is generally more dovish than Powell is accurate here. But it isn't always accurate over the course of the last year. If anything, if we average the last six or seven months in terms of this indicator that we put together here, you'll see that it's fairly close to neutral. The view that Brainard is going to change the way the Fed is run, change the DNA into a more dovish perspective, is not necessarily being borne out by how media is covering it. I would look to another strong reaction, where we saw the short-end sell-off and potentially the curve flatten as a potential opportunity to fade that view. If anything, media is showing that a Fed under Brainard might not be that different than a Fed under Powell. With that, I think I'll conclude and hand it back to Meg.Well, thank you Ronnie and Marvin. We'll now open it up to Q and A from the audience. As a reminder, you can continue to ask questions via the sidebar, so we'll open it up now. I think the first question I'll direct to you, Ronnie. We talked about how Chairman Powell has market impact just through his own - the coverage of his tone, or the tone of his coverage. Are there other individual voices that came through as also having, I guess, meaningful market impact beyond Powell?It's a great question! We haven't looked at that; I think Marvin and his team might have looked at that more carefully. The issue is, when you get into the individual members, the coverage is more sparse. So it's a little bit more difficult to run a regression and to do some meaningful statistics with that measure. Marvin and his team might have looked at more anecdotal evidence in more recent times. Marvin, do you have anything to add to that?Yes, absolutely. Powell's voice certainly rings loud. I think that the normal gaggle, if you will, of most active speakers within the voting members has had a bigger impact. Within that group I would include Brainard, I would include Clarida. Some of the more active voices within the regional presidents don't seem to have as much of an impact. The market certainly is focussing on certain people. I'd love it if Ronnie and his team did some more statistical analysis rather than me just lining up their lines with other lines in the market. Certainly we have seen more market reaction from those people that I think carry a bit more weight in their voice.I guess to add on to this, I know that a lot of the work that you do at MKT is correcting for biases and accounting for biases that we find in media coverage. In the work that you've done - and maybe I'll direct this again to you, Ronnie - have you found any biases in terms of the tone for coastal versus central members, or for male versus female members? Are there any specific biases you've found in your work?It's a great question! Again, I welcome these types of questions; it gives us more things to look at. The biases that we've seen, the way we calculated the indicators, we were looking at the hawkish minus dovish divided by the sum; that seems to correct for many of the systematic biases across sources and reservoirs and also across time. That's what we were more worried about because what happens during the meetings, there is much more media coverage. So the concern is that you might have a different number of articles, but not only in that; it's just that the tone will be systematically different during meetings. So we try to correct for that by normalising by the sum. We have shown that that actually helps quite a bit in normalising the overall level across different periods. The other suggestions about biases are well taken and we'll certainly look at them.Someone says: great work, and they're curious if you've worked on similar analysis for developing countries, in particular in Asia and Latin America.The sixteen banks that we cover of course have some Asian - the Bank of Japan, of course, is in it. We haven't done yet a full analysis of the more developing markets. I think it's a good suggestion. What we need to do is to make sure we have enough coverage. The distribution of coverage is highly skewed towards the more developed countries. I see where you're coming from, it's a good question, and up to us to go back to try and look at that. If we have sufficient information then we'll certainly produce these indicators.Marvin, I think I'll direct this one to you. At the end of the presentation you talked about Fed vacancies and how that impacts hawkish-dovish views. What are some of the other Fed vacancies in the administration? Would that affect some of the comments that you have made?Yes, absolutely, it's going to be a big topic. We did have a few Fed presidents resign recently because of some of the trading activities that occurred. So there is a bit of a filling that needs to go on in Boston as well as DC. I'm sorry, Boston and Dallas. There is a vacancy that hadn't been filled by the Trump administration on DC. Powell versus Brainard: potentially if that moves, how do we replace Brainard? We've got Quarles, who said he's going to leave at the end of the year. It's still questionable if Clarida will stay if he's not the vice chair. The way I would approach it really is to look at how media coverage, each one of these individuals, and to think through what the administration and or the regional bank boards will do in terms of replacing them. It's essential, it's super important just because the votes on the dots are so close within 2022. There is so much focus on what the 2023 outlook is that this is good work that we all should be doing now. I'm actually working on a few things with Ronnie's team to suss all of that out. We're going to hope that we hear what the administration plans on doing for the Fed chair first. There's a lot more work to do beyond that.I guess another question - and you started to talk about this I think, Marvin, with some of your comments around the ECB. We just had a series of central bank meetings. Some of them were surprising, at least in the case of the ECB. I guess do you have any more comments around what sort of insights these indicators would've helped provide around the recent bout of central bank meetings and announcements?Most certainly the ECB, sounding hawkish from the media's perspective is in and of itself pushing against the norm, if you will. Bund investors, EGB investors would have benefited by really taking a hard look at what media is saying with regard to how the ECB is approaching things. The BoE has been fascinating because it certainly has moved far more hawkish in terms of what they're saying. But from the media perspective it's moving more sideways, which is corresponding to the choppiness that we're seeing around the views as to whether or not the BoE can actually have rate hikes as the market is pricing at this point. That's a story that is being played to again and again and again across not only the developed world but the emerging world as the central bank dispersion, if you will, is more than we've seen certainly since the pandemic began - but well before the pandemic, even. So there's a lot going on and this is again just another alternate data source that I use fairly aggressively to help me sort through things.Well, we're at time, so thank you again, Ronnie and Marvin. This work will be really valuable as we navigate central bank policy going forward. Thank you for your time.