Will Kinlaw: Thank you, Michael. So as Michael said, this talk is really about how narratives can drive. And it's actually a paper that is not done yet. So you're the first audience to to see this work. You can see the co authors listed there, my colleague Zachary Crowell Leverage, and Michael Guidi, as well as Ronnie Sadka and Gideon Ozick, who are at media stats. So Ronnie teaches at Boston College. He's going to speak later on how narratives impact pricing and the equity markets. But this paper is really about narratives and currencies. And as Michael said, it stands to reason. We know that in all of our conversations about currency is whether they're watercooler chats or Bloomberg chats or meetings with strategists. We talk a lot about narratives and the narratives kind of come and go. And really what we're going to try to do here is to quantify some of that and see what it can teach us, not only about what's going on right now in markets, but also what might be going on in markets in the future. So for a little bit of inspiration, this is a quote which I won't read to you from the Robert Shiller book, Narrative Economics, which came out a few years ago. He actually spoke at this event when it was virtual during the pandemic a few years ago. And it was a great talk. If you haven't read the book, I highly recommend it. It gives a really nice history of how narratives have impacted economic events for the last few hundred years, and it has a lot of great anecdotes in there as well.

In fact, one of the topics that came up this morning of Greedflation gets a lot of play. There's nothing new about companies being accused of taking advantage of inflation to gouge consumers. But perhaps my favorite narrative from from this book had to do with capitalism versus communism. And it was a joke that went viral in the 80s after it was told by President Ronald Reagan. And the joke goes like this There's a man in Moscow who decides he wants to buy a new car. So he goes to the car dealership and he picks the car he wants and he goes to the salesman. He says, This is the car I'd like to buy. Can I buy it, please? Salesman takes the money and says, Certainly, you can come pick up your car exactly ten years from today. And the man says, Oh. And he pulls out his calendar. He says, Morning or afternoon. The salesman says, It's in ten years. Why does it matter? He says, Well, the plumber is coming in the morning. So an example of a joke to do not with, but with with capitalism and communism that went went viral as a narrative and impacted the way people think about about markets. There's many other great examples in there. So there are three key conclusions that I want to talk about today.

The first is that we can quantify narratives, the amount of coverage, the tone of coverage, using the technology that our colleagues at media stats have developed, and that's available to you on Insights. The second is that currencies exhibit a lot of narrative exposure. It's time varying and it varies across currencies, but there's a lot of significant exposure to narratives. To me, that's the most interesting point here. And the third point is that narrative exposure is, it appears, can actually help us forecast returns alongside institutional flows. So let me take you through some of the evidence. And I'm going to start with how do we measure a narrative? And this this will be a review for some of you who followed our research for a while. But I think it's really important to establish how we measure a narrative before we can talk about how narratives impact currencies. So for this paper, we have a universe of 103 different narratives. I'm showing you 18 here. The narratives are all economic and financial narratives, and it stands to reason could impact currency markets. The way that we measure narrative attention is that the media stats team has multiple reservoirs of articles. So these are all media articles. It's digital media, it's major publications like The Wall Street Journal, its press releases, its blogs, its industry publications. So it's over 100,000 sources a day that are being monitored, scanned, using textual analysis and captured. Now, the way that they measure attention to a particular narrative is as follows.

First, they have to identify which articles are relevant to that narrative, and that's done by looking for certain keywords and phrases that relate to the narrative. Obviously, if you have an article that's talking about recession, you can link that to the recession narrative. But it's more complicated, right? It depends on a narrative that may not be directly discussed. You have to use certain phrases. So identify articles that are relevant. And then the next step is to give each article a sentiment score. So that score tells us, Is this a positive article or is it a negative article? And that has to do with looking at the words that are used and the phrasing to understand the tone of the article. Now, what we use to measure narrative attention is what we call negative intensity. And I'm not going to go into all the details, but it's the second equation here. And what negative intensity does is it counts the proportion of articles about the narrative that have a negative tone. So it's one measure, but it tells us two things. It tells us about the amount of coverage, about the narrative, and it tells us about the tone of the coverage. And both those things are very important, right? We need to understand what is the direction of the sentiment. We also need to understand what's the volume of the coverage. So that's what this measure does for each of those 103 narratives.

And it does it every day. So just to give you an example, I've pulled out here three different narratives. And Gideon Ozick, my colleague who's here, went through and looked at each narrative and how many articles we actually had in the reservoirs for this time period going back to 2015 for each narrative. So you can see it's a very large number of relevant articles. And as you can imagine when you're talking about a narrative like recession or inflation, most of the articles are negative, right? You might have a positive article about a recession where it's saying maybe recession fears are easing, but most of the time the articles are negative. What we're interested in is the changes through time of the amount of coverage and the tone of the coverage. Here are examples of selected articles from the time period for each of the narratives. So you've got there the headline and you've got a snippet of text that was relevant to the to the AI that's doing the reading of the article and flagging it. So you can kind of get a sense for the types of articles that are serving here as inputs. So I'm going to show you a couple examples of what these negative intensity series look like for different narratives. And I'll start with recession. So you can see this one was pretty quiet for most of this period from 2015 to 2019 and 2019, it started to pick up.

We saw an inversion in the yield curve. We saw weaker industrial production, and you started to see discussion about could there be a recession even before Covid arrived on the scene. Interestingly, one of the articles that's captured here, this one about Brainiac's adapt human skull analysis to predict recessions. Mit and State Street Researcher Index forecasts over 70% Chance of Recession was actually an article about a paper that Marc and Dave and I wrote about business cycles and predicting recessions. I like doing papers with them because I get mistaken for for being a brainiac. But that was late 2019, early 2020. And then, of course, Covid hit and you saw a big spike in discussion of recessions and indeed did see a recession during Covid. Now, then things calmed down a lot. And recently, as we saw the interest rate cycle start to change, you saw more discussion of recession narrative. So you can see how there's a lot of variation in time here on the on the negative intensity measure for the recession narrative. Here's the inflation narrative looks a little different, right? Inflation really wasn't an issue that was discussed much until 2021. So and then it peaks in the middle of 2022 and starts to ease off. And here's another interesting one. This is the trade war narrative. So this is a narrative that, at least for now, has had kind of a defined life cycle in 2018 and 2019.

So you can see it kind of peaked. We saw a lot of discussion about trade wars during the Trump administration, and that's really tailed off in the in the few years since then. So it gives you a sense for three different major narratives that clearly have implications for economies and for currencies and how we see this time variation. I'm not going to show you these charts for all 103 narratives, but hopefully it gives you a sense for how these series behave. So what we're going to do now is try to measure the exposure of currencies to changes in these narratives, and we're going to do that in the simplest way possible. So this is a univariate regression. We're regressing. One week spot returns. We have 52 currency pairs, we have all the main dollar pairs, and then we have a bunch of cross pairs and we're going to look at is there significant exposure of these one week returns to one week changes in negative intensity? And we do some normalization which is described there in the in the footnote. But that's the idea. It's very simple. And the punchline is there's actually exposures everywhere. We see a ton of significant exposure, but it varies a lot. Again, it varies across currencies and it varies through time. So I'm not going to again, I'm not going to show you there are thousands and thousands of regressions here, but I'm going to show you a few charts to give you a sense of what we see out of those regressions.

So this chart shows for each currency that we looked at the average number of narratives each day to which that currency has a significant regression exposure. So do changes in the in the negative intensity of that narrative relate or correspond to changes in the currency. And what you see here and we define significant by a t stat of greater than 1.5 or less than -1.5. We could dial that however you like. But the point here is that you get a lot of exposure, much more than you'd expect purely by random chance, and it tends to be somewhere between 20 and 30 different narratives that a currency may be exposed to at one given time. Here's the same view, but from the perspective of narrative. So these were the narratives, the 20 narratives in our data set that had the most currencies exposed to them with some degree of statistical significance and unsurprisingly, economic condition. Narratives pop up a lot. You see Covid in there, you see commodities, you see risk in the financial markets. All kind of macro narratives that we would expect would influence currencies. So let me show you a few examples here that that show a very specific time periods. So I'm going to show you four charts and each one is going to zoom in on a particular time period where we know that a narrative was important to a particular currency.

So the right hand sorry, the left hand side of this slide looks at the Chinese yuan during the trade war period and what this chart is showing, the shading, red shading indicates that during that time period, the currency had a negative statistical exposure to the narrative. So negative is a little tricky here. When negative intensity goes up, it means that the currency is going to go down, right? Green shading means there was a positive exposure. So a negative coverage, the narrative goes up, the currency goes up. And we're quoting just for simplicity here. We're quoting all currencies in dollar terms. So it's the dollar cost of the foreign currency. And the orange line shows the actual negative intensity series. So that's the the narrative measure we're looking at. And the blue shows the currency. So you can see here for the one, you see a very clear period of negative exposure when negative coverage of the trade war goes up. That's bad for the one. And you can see that indeed during that period you see a depreciation when the negative intensity measure spikes. Now, the other side of that coin for the trade war is the yen. So Japan obviously has a very large amount of foreign invested assets. When things are looking risky, a lot of those repatriate which drives up the currency, it's a safe haven currency and you see that here as well. You see a lot of positive exposure to the same narrative and you see that when the same narrative spikes, you see an elevation in that currency.

Here are a couple of other examples. Obviously a classic case would be looking at the pound and Brexit. You can see here on the left, the pound had significant negative exposure to negative coverage and the Brexit narrative. And indeed when when that coverage spikes, you see a depreciation in the pound. This one on the right is really interesting. So a few weeks ago I called up Lee Farage and I just said, Hey, what narratives are being discussed in a lot of your meetings with investors? And he was on the tarmac, I think in New Orleans, probably talking to the passenger next to him about demographics and economies or something. And he said, actually, what keeps coming up is onshoring this notion of companies bringing workers and functions closer to home and out of hot places that could be international hotspots. And I said, okay, interesting. We have an onshoring narrative. I said, What currency do you expect would benefit from the Onshoring narrative? And he said, Mexico. And indeed what you see over the course of 2023 is that the Mexican dollar has this positive exposure to the onshoring narrative. So you can go on and on. There are a lot of these we had to pick. We had like 20 of these charts. They're all pretty interesting. They kind of give you a sense for when does a currency have exposure, narrative y and which direction is it? So what I want to do for you is to quantify a little bit the economic magnitude of these effects.

What this chart shows is very simple. I've shown you that we can measure the exposure of currencies to narratives. We can determine whether that's significant by assigning some thresholds. So we remove the noise and we can determine the direction of that exposure. What this shows is when you have a shock to negative intensity, either negative intensity falls for a narrative or it spikes. Do the currencies that have positive exposure behave differently than the currencies that have negative exposure and you're measuring the exposure in the prior period. So this is helping us understand, is narrative exposure persistent or is it just kind of jump around randomly from one period to the next? And what this chart shows is what you'd expect when you have a bunch of currencies with positive exposure narrative. And that narrative spikes those currencies, outperform currencies that have negative exposure to the same narrative. So I think it's important in establishing that these effects are persistent through time. We can use a recent historical sample to figure out which narratives occurrences are exposed to and the degree and direction of that exposure. So why is this interesting? I think there are probably a number of reasons. And I've talked to a number of our traders about this. Obviously, certain narratives, we all understand what they're going to do to certain currencies.

But what this really lets us do is ask the question for a given narrative which currencies have positive and negative exposure? How could we build a basket of currencies to track the narrative? We want to make a bet on the narrative. How could what would be the best basket to do that? That's balanced. It also lets us take an existing portfolio, be it a currency portfolio or any kind of portfolio, and ask the question to which narratives is this portfolio exposed? If we see a shock in these narratives, what do we think is going to happen to this portfolio? What would be the direction? What will be the magnitude? So here's an example of that. This is the recession narrative. The top three positive exposures are shown on the left and the negative exposures on the right. So you have three safe havens, at least now, safe havens on the left and three kind of high beta commodity type currencies on the right. So it makes makes good sense, but this gives you a way to kind of quantify and it gives you a sense of how you might build that basket. And I don't have time to get into all of it today, but if you actually build a time series for a recession basket of currencies, it tracks the recession narrative reasonably well. All right, so let's bring flows into this. Flows are another important dynamic that we measure.

Of course, looking at the institutional information that we have at State Street and understand how flows and narratives interact and how they might help us in forecasting returns. So we specify here a very simple, again, regression model. We're going to try to forecast the next period return. We're actually going to look out for weeks and we're doing that using two variables. One of them is the change, the past change in the narrative. So the negative intensity for the narratives that that currency is exposed to. So we're not going to look at all narratives. We're just looking at the ones that that currency that have been important, that have been associated with that currency, with a degree of statistical significance in the past period. And we're also going to look at the flow in the past period in the currency, and we're going to attempt to use those two things to forecast the return next period. Now what we do is we take the negative, the narratives, the cases where a currency has negative exposure in the narrative, and we just flip the sign. So we expect all the narratives to have a positive association that just makes things a little simpler. We can do this with one variable instead of two variables, and here are the results from those regressions. So again, this is using historical data to forecast out one, two, three, four weeks what is going to happen in the return of the currency using flow and narrative information? And there's a couple of things here I want to point out.

The first is, as you expect, flow is significant in the first week. After that, you have a bit of a reversal effect. This has been well documented. If you want to understand how flow dynamics work with currencies. Alex cIma Fox has a great session coming up to do that. One of the things that we've done here is to also condition on holdings, which really improves If you have a situation where investors have a strong overweight position and start selling, you can have higher conviction than if you're just looking at the flow alone. But the point here is to really focus on the narrative exposure and just make sure that it's giving us some incremental information that we're not getting from the flow. And that is indeed what you see in these results. You can see the T stats are quite significant for all but but week three, looking forward and again for week four here, we're still using information now that's four weeks stale and we're seeing that the positive move or negative moves in the narratives that are associated with that currency statistically seem to persist and seem to give us some momentum that can be helpful in forecasting what's going on with the currency going forward. So again, just to give that some some kind of some economic magnitude for you. This is the same chart I showed before, but now we're looking not at the same week, the contemporaneous week, but the next week.

So if you have a week where you have a strong spike and the narratives that a particular currency is exposed to, what do you see for the positively exposed versus negatively exposed currencies in the following week? And it's exactly the relationship that you'd expect, particularly when you have a negative shock. So a negative intensity is is falling. So let me just recap here some of the key conclusions. Number one, we have this technology to measure market narratives and not only measure the volume of coverage, but also measure the sentiment, the tone of the coverage. Those are the negative intensity series that I showed you. And we do this now for 103 narratives, and we're always adding to it. In many cases, it's because one of our clients calls us and says, Hey, do you have a narrative for X? And we'll go and look and add that narrative. The second point is that we see a lot of significant exposure of currencies to these narratives. They vary through time, they vary across currencies, but it's there and it's something that could be helpful to track. And I'd like to point out on this on this point. We looked at, again, 52 pairs. We've looked at the dollar pairs and we've also looked at all the non dollar crosses. And the results are quite consistent. So it doesn't seem like there's an effect here that is just being driven by the dollar.

And lastly, we see some out of sample evidence that the narrative momentum or a narrative factor can help us forecast next period returns. And that seems to be separate from what the forecasting power that you might get from institutional flows. So I'm sure one of the things that you may be wondering is what's going on right now. So I'll take a vote, please. Which three narratives right now have the highest degree of negative intensity. So the largest volume of negative coverage right now is that the labor market, inflation and recession, is it equity investing, inflation and recession or is it equity investing, Inflation and interest rates? So it looks like the audience got this wrong. I also got this wrong. I thought it was going to be number one. It turns out the answer is actually number two, it's equity investing, inflation and interest rates. And if you're interested in knowing which currencies have significant exposure, positive or negative to these narratives, there's quite a few. I don't remember them all, but my colleague Zach Crowell, who's here, has them on his phone so you can find during the break and he'll tell you or we can email you which currencies are exposed. But there's a package of currencies that seem to be exposed positively or negatively to each of these three narratives that we see spiking right now. So I'll stop there and I'm happy to take any questions, if there are any.

Speaker2: So I've got a I got a couple on here when I start with here and then we'll see if there's any in the room. So were you able to measure a narrative that relates directly to factors like Carrie?

Will Kinlaw: Interesting. So a couple of things on that. We actually do have a suite of sentiment indicators that relate directly to currencies. So that's looking at what is the sentiment toward the euro or the pound. And those are helpful. But obviously when you're talking about that, you're dealing with a smaller set of articles. Our goal here was really to understand how the discussion of these broad macro factors that impact the whole market spill over and impact currencies. We have looked at factors and I'm looking at Gideon because I don't know if we've looked at Carrie, but we do look at things like value and momentum and equity type factors. So we have started looking at that and we'll continue to look at that, but we don't have results on that here. Ronnie is going to talk about how factors can impact equity pricing. So there'll be more discussion of that this afternoon.

Speaker2: Okay. Any hands? I've got lots on here. So let me let me go through some of these market movements can shape the narrative. So curious if it would be beneficial for us to factor in the direction of causality between narratives and markets.

Will Kinlaw: So yeah, that's interesting.

Speaker2: Circular. That's a good question.

Will Kinlaw: Yeah. So we did look in some regressions at including past returns. You don't get a tremendous amount of significance there, which wasn't surprising, but probably a lot. My suspicion is that the market does impact the narrative. But what we do see is that there is a period of time before a narrative gets fully priced in. And it seems like that even though this is all public information, the way that it's being captured and aggregated and organized is adding some value. And I think it's akin to what Peter said on the panel and thinking about Geo Quant and geopolitical risk in emerging markets. It's the same kind of challenge. You're trying to scan massive amounts of information and come up with some summary information that's going to help you understand how it relates to a particular country or particular narrative. And it seems like that's not all getting priced in right away. At least that's what these results suggest.

Speaker2: So this is another really good one actually. Have you tried any topic modeling or unsupervised clustering techniques to discover narratives investors might have missed?

Will Kinlaw: No, that's that's a great question. We haven't and that's something that we're thinking about. And if you another session I'm going to plug is Gideon and Jana Minh tomorrow are going to have a whole session on large language models. And that's the same modeling that powers this work, but not just how we use it here, but also where they see that space going and implications for investing. So I think they're more qualified to answer that question than I.
Speaker2: We'll go there and ask it. Gideon Okay. There's one from the floor. Here we go.

Speaker3: Oh, hi. I'm wondering, what kind of technology are you using in analyzing the narratives and also which language is for articles are you able to analyze?

Will Kinlaw: Yeah. So right now we're covering and I don't know if we can get a mic over to Gideon because that technology question, I'm definitely not answering. But, but from a language perspective, we cover English language media, but it's English language media from all over the world. And obviously if you had to pick one language, that would be the one. But I'll see if Gideon wants to talk a little bit from a technology perspective or correct what I just said.

Speaker4: No, you were right. So in terms of languages, we actually collect languages above and beyond English. And one of the things we'll talk tomorrow is how you can use kind of advanced techniques in language modeling to actually go beyond beyond English. We've had a few specific examples or use cases brought by strategists where we looked at Italian and Spanish and Portuguese to analyze very nuanced type of scenarios in the local markets, reading specifically information from that, from from this local in these local markets published by local newspapers.

Will Kinlaw: So we'll continue to work on this. This is really the first kind of formal effort to quantify it from an FCS perspective, and we'll continue to work on it. So please stay tuned.

Speaker2: Another these are all terrific questions. Can narratives, can narratives predict themselves? I.e., is there a momentum in narratives and can they predict macro? So does talk about inflation predict inflation?

Will Kinlaw: Huh? That's interesting. So there is some positive autocorrelation in the narrative itself. Interestingly, and this isn't the question, but I think it bears pointing out the the cross-sectional correlation is lower than I would have expected. One of my first questions was, is there some like sentiment super factor in all these narratives are just kind of a little bit of noise around that that's not the case. The average correlation across the 103 narratives is only 15%. And I think the standard deviation was like 13%. So there's definitely a lot of variation in what these narratives are telling us. We do see some positive autocorrelation in the in the narrative series. The question of do narratives predict inflation is fascinating. I we haven't looked at that per se. My guess, though, is that obviously most market based variables predict inflation. If you're talking about CPI, right? I mean, break even spreads or most market based variables, even consensus forecasts are going to lead CPI because CPI is such a lag associated with it. So my my intuition would be that this would probably lead CPI, but but probably not something that was much higher frequency like price stats. Indeed, you'll know this better than me, Michael, but if you look at that, maybe we can go back to that one slide on the the inflation narrative. I think price stats started picking up before the inflation narrative really took off.

Speaker2: Yeah, no, I think that's right. And actually, there's a couple of joint pieces that Gideon and I have been doing on supply and how that relates to inflation and particularly looking at whether corporates are mentioning labor shortages or shortages of goods, which absolutely dovetails with Alberta's work on measuring shortages. So there's a lot of overlap there and a lot more scope for us to do more in that vein. Okay. I think we probably have time maybe for one more question. Do you have a library of narratives that you monitor each week to check the changes in what is driving currency moves? Perhaps develop an eye feature and insights?

Will Kinlaw: Yes. So the answer to that question is partly yes. So we do have a library of narratives. We are always adding to it. It is on insights. That's all there. Moreover, our strategy and it's by the way, it's not weekly, it's daily. And our strategy team is constantly looking at that and pulling out what they think is important. And I think where we'd like to go with this research is to develop some new visualization tools to figure out which narratives are flashing and what does that mean for which assets. And that's absolutely where we're we're going to go next.

Speaker2: Okay, brilliant. Well, I think that was a terrific demonstration of a set of questions that give us our research agenda for for for probably many, many months and years to come. So anyway. Well, thank you very much.

Will Kinlaw: Thank you.