US-China Tension: Using Text As Data In Policy Analysis

>> Steven J. Davis: Welcome to the Hoover Institution workshop on using text as data in policy analysis. My name is Steven Davis. I'm a senior fellow at the Hoover Institution, and I organize this workshop with Erin Carter. Today we are delighted to welcome Bo sun, who will present her paper on US China tension, co authored with John Rogers and Tony Sundae.

She will speak for about 30 minutes, followed by about 25 minutes of Q and A. Then we'll turn off the video recording and have an informal discussion for anyone who wants to stick around. Boson is on the faculty at the Darden School of Business at the University of Virginia.

Before joining Darden, she was a principal economist at the Federal Reserve Board of Governors. She also served on the faculty at the School of Management in Peking University, among other positions. I've known Bo for several years and she's a good friend and I know her to be an excellent wide ranging economists, and we're going to hear about some of her great work today.

Bo, we're delighted to have you.

>> Bo Sun: Thank you so much, Steve, and thank you all for joining me to discuss us China tension. Let me share my screen here. So, as Steve mentioned, this is joint work with John Rogers and Tony Sun. Tony is research assistant currently at the Federal Reserve Board and is applying for grad school this year.

The usual disclaimer applies. This is our view entirely, not the view of the Fed. Just to briefly motivate what we do in this research. As we all agree, the US China relation has undergone profound changes over the past few decades. And if we just focus on the recent few years, tensions have escalated across a broad range of issues.

And these two countries, they practically compete in virtually all areas, not just economically, but also in terms of political influence and technological dominance. And also, as we all know, these two largest economies are also very much intertwined with very deep trade and business ties. So all of that makes their rivalry really complex and consequential.

To that end, the rising tension has been cited as a key determinant of economic activities by market participants and also by policymakers. But to many others, though, such concerns just seem perhaps hyperbolic and maybe even all right, wrong. So does the rising tension really matter for economic outcomes?

We think it does, but we have a lot to learn about how it does. Although the bilateral relation has garnered a lot of attention lately, its significance in driving economic outcomes has now been empirically studied in the context of a long time series. Outside the particular trade war episode, a binding limitation has been the lack of a quantitative indicator, an indicator that can quantify the intensity of such tension and perhaps track its fluctuation over time, and an indicator that's consistently available through time.

And that's precisely what we turn to in this research. So here we primarily ask two broad questions. The first one is, can we quantify public concerns over us China tension? Can we actually come up with a robust metric that covers a wide and consequential set of issues? A set of issues including trade conflicts, but also extending beyond those?

And then the second question we pursue is that does this rising tension really matter materially. Does an escalation in the bilateral tension really have a tangible bite in any of the real decisions that firms have to make? Are there any significant effects showing up in the aggregate? A very quick preview.

In this research, we offer a numeric indicator of the US China tension. And I'm going to show you today that our metric shows very close alignment with the related views expressed by business decision-makers and also by policymakers, both in terms of what they convey in words and also what they implement in actions.

And then with this index that we construct and carefully vet, we do find that escalation detention does have significant effects. So at the firm level, we find that rising tensions would be associated with protracted declines in firm investment. It's also associated with us firm supply chain reconfiguration away from China, and we also find negative transmission in the aggregate data as well.

More importantly, I want to share upfront that what we find is that these contractionary facts of us China tension primarily operates through these uncertainty channels. So it's really the fear of escalated friction that's driving a lot of the economic responses we see in the data. So one way to look at what we find is that we really don't need tangible frictions being put in place, tangible barriers being under being put in place to see the economic ramifications playing out.

The mere thread of those, the fear of those frictions would be sufficient to generate economic facts that are quite adverse. Put it differently, though, another way of interpreting our results, another way of freemium what we find. Is that it's really depending on whether and when economic decision makers are going to become largely set on a new norm, when they fully factor in the new norm.

Now independent of the level of the tension that would characterize the new norm, as long as uncertainty updates, much of the economic effect might become rather contained then. Now, starting from the measurement, this notion of US China attention, it's really massive. It's very much textured, not obviously amenable to quantification.

Why do we want to engage in this labor intensive and honestly very time consuming endeavor to give it a numerical reckoning. What I will share is that our belief here is that what gets measured gets analyzed. There is a sense in which, once we have a numeric indicator that could give us a sense of the magnitude of the tension and how it has fluctuated over time, it would give us some basis for characterizing what has happened over a long period of time.

Hopefully it would also give us this means for studying, for analyzing what the economic ramifications might be. And here I also want to take a step back and situate our pursuit in this broader context of this evolution of the bilateral relation over time by zooming out over the past several decades, the tension has.

Has gone from mostly ideological or geopolitical, if you will, towards economically focused. And if you just zoom into the recent few years, the tension has been shifting from trade war towards investment war, tech war and national security conflict. So one motivational perspective of this project is that I think it's important for us to recognize that US-China tension has multiple facets.

It actually covers a broad range of issues that are also dynamically evolving. And so we have fantastic papers that study the economic effects of the trade war that started in 2018. But I think it would be also important to have a numeric indicator that can take seriously this complexity in the tension that would allow us to capture some of the multidimensionality to it.

Now, on that note, I wanna be clear on what we are trying to capture here, so what we aim to capture is this public concerns over US-China attention. It's a perception that the public, by and large holds, including households and business decision makers. To get to that, it takes us to study news data, and I will also say that this news petrol data also offers this vehicle that allows us to capture some of the multidimensionality to the tension.

There are broad sources of the tension, now familiar to all of you on this call. Steve has this massively influential paper on economic policy uncertainty, together with Nick Bloom and Scott Baker, where they practically pioneered this growing literature on newspaper analysis. The premise of that literature is the newspapers have got to reflect readership, if they don't reflect readership, they won't sell.

And there's also additional angle to it, which is that newspapers can also shape the perception of their readers. And their readers should includes a large set of business decision-makers, even including the most influential. So with that thought in mind, we follow the general methodology laid out in Baker, Bloom, Davis 2015 QGE, and use the media coverage.

Quantify the media coverage on the issue, on US-China tension, and use that as a gauge for public concerns over US-China tension. So basically, we just compute the share of newspaper articles containing keywords from these three broad categories, UCT. So for a newspaper article to move up our index, it has to mention the US, it has to mention China.

It also has to contain keywords indicating tension on at least one contentious issue. I will say that here, to take into account this time variation in media coverage on China, we are scaling the raw UCT articles, if you will, by the total number of articles mentioned in China in the first place.

Now we do this automated search in this list of major US newspapers, including the Wall Street Journal and New York Times. And so these are widely circulated newspapers that business decision-makers also likely follow. So this search based methodology really allows us to update our index in real time, which we consider crucial, and also it allows our index to be easily interpretable.

Now, so that's what we do mechanically here, I just wanna spend one minute conceptualizing what really constitutes US-China tension. Right, what in principle should we think about US-China tension, the public concerns over that, independent of how you might wanna exactly measure it. So first and foremost, I think rising tension could in principle imply new frictions, new tangible barriers being put in place that would abduct the normal economic transactions between the two countries.

So tariffs would spring to mind, but we could also think about, for example, the expanded oversight over American investment in China, the executive orders that increased scrutiny of American funding of Chinese startups and tech companies. We could also think about these policy changes that make it extremely difficult for Chinese firms to be listed on the US stock exchanges.

And over the past decades, especially recent years, we've seen firms being blacklisted, sanctions being issued. There are also restrictions imposed on both sides that would limit technology transfer and share know how. So all of that would be tangible actions, concrete barriers being put in place. Now, I would say that on top of those, or even in absence of these, rising tension could also trigger uncertainty about escalated friction going forward.

Uncertainty surrounding these arguably principal instruments of bilateral engagement. So, when we think about US-China tension, these two components, the realization of tangible barriers and their surrounding uncertainty, these two components are inevitably blended together. So as a measure that quantifies public concerns over the issue, our measure is gonna be a holistic measure.

It's a holistic measure, both in terms of the range of issues that covers and also in terms of combining these two components together. I just wanna make that clear, now to back up a bit, how did we come up with these search terms in the first place? In this research, instead of relying our own judgment, we are relying on machine learning algorithms to help us decide what words are likely relevant, what phrases are most frequently used in newspaper discussions on the issue.

So the way we go about it is that we started with the articles that mentioned the two countries in the first place, and we randomly selected 5% of these articles, and we worked with research assistants to flag those articles that discuss US-China tension. So that process gave us about a thousand newspaper articles that are manually classified as discussing high were rising US-China tension.

And then we proceed to perform classroom analysis using the three commonly used algorithms k means LDA and news map non surprisingly, there is a lot of overlap in the output we generate from these three methods. And we take an expensive approach and group all the interpretable terms from this exercise and group them in the three broad categories that I just showed you UCT, US-China intention.

In our category terms we have phrases indicating confrontation and we also have phrases indicating specific contentious issue. In this research we did a medium scale on the smaller side, human audit with the objective of refining the sort. Search terms, the required search term refinement turned out to be fairly minimal.

We did away with the Chinese leadership last names to minimize the impact of false positives, and we also got rid of words including bar and virus, but that was pretty much it. So here, let me show you our indicator of US-China tension. It's plotted here at the monthly frequency.

The series has been normalized to have an average of 100 over the sample. We take a lot of comfort in seeing that the index fluctuates in ways that are in line with the related narratives. So you would see that the index spike up around the Belgrade embassy bombing and then the spying standoff in 2001, and then it stays elevated throughout the trade war period, spikes up around the Huawei executive arrest.

And it also shows escalation at the onset of the pandemic for mutual blaming over the outbreak of the virus and at the onset of the war in Ukraine. And there is the recent spike last year over the balloon incident. Now, actually once through every single spike here, large and small, all of them could be pinpointed within minutes to a fairly salient event that happened in the months.

And I also cross checked these with a number of commentaries on the bilateral relation, including the ones produced by the Council on Foreign Relations Reuter and a pretty comprehensive survey done by the Congress. It's by no means a test of the index informativeness. I will tell you how we exactly value the information content of our index, but the alignment is reassuring and it's in part reflected in the spikes that you see on this chart.

Now, on that note of evaluating the information content, so we know that there is the growing literature on news based actual analysis. And folks have made long strides in showing various proofs of concept that newspaper can be this useful vehicle for the spread of ideas and in terms of reflecting, reflecting and also shaping public opinion.

Nonetheless, we wanna look to see if all of that applies in our particular context as well. So we've done extensive exercises for the purpose of evaluating the informativeness of our index. I'm gonna show you some of what we've done because they really boosted our confidence that our index can serve as a useful signal.

Now, let me start by showing you that our index can be a good gauge for business decision makers perception over the bilateral tension. To get to business decision maker perception, return to US public firms primary communication device, which is this quarterly earnings calls that they hold with interested parties where firm management would discuss important aspects of firm performance and decisions.

So following the idea, the methodology in Hasan et al 2019 QGE, the way we do this is to calculate the share of earnings call discussions devoted to the topic. So specifically, we count how many times China gets mentioned in close proximity to any words indicating tension. So we could just count how many times the word China or Beijing would show up within ten words apart from any of our category T terms.

And we subsequently scale that by the length of the earnings call transcript. So we do that for every single earnings call a US public firm held, and for each quarter, we average that out across all firms. That would give us this aggregate time series, which corresponds to the average share of earnings call discussions devoted to the tension.

And you could see that this earnings call base, the UCT index, if you will, and this has also been normalized to have an average of 100 over the sample. It tracks our news base, the UCT, fairly well. Now, the next set of exercises I'm gonna show you, it's done in the context of firm real decisions.

If you think about why firms would be concerned about US-China tension, what makes US-China tension salient economically it's that it raises both the likelihood and uncertainty about escalated friction going forward. So if we think about it in terms of investment payoffs from firms perspective, there is a downward shift in the mean, and there's also an increase in the mean preserving spreads.

And they both have the same directional effect, escalated friction the way on investment payoffs would naturally discourage investment. And as we all know that based on the real options theory, the uncertainty being engendered here could also create a wait and see type of attitude and generating incentives to delay economic decisions that are costly to reverse.

So to that end, we would expect rising tension as registered as an escalation, or UCT, would be associated with a retrenchment in firm enhancement. And we would also expect that effect to be particularly pronounced among firms likely impacted by the tension. So I will say that this set of analysis, we pursued it for the purpose of evaluating whether our index correlates with firm decisions in ways that make economics sense.

But I think this set of exercise also is gonna help us have a sense of how public concerns over the tension could get propagated through the real economy. Now, let me show you the simple regression we run first. This is the average association between UCT, our index, with firm investment.

So we're just putting in lagged UCT in otherwise standard test of Q theory of investment. The subscript l here, it's the quarter lag between UCT and the investment. So we're estimating how a rise in UCT today affects from investment next quarter up to four quarters out. The main dependent variable is this investment rate, which is capital expenditure scaled by lag of total assets.

We have all the standard controls on the firm side, as well as controls for current and expected macroeconomic conditions. So a negative beta would be indicative of rising tensions delaying firm investment. And that's what we find. And the effects are significant up to four quarters out. And it's also robust at industry level and robust to using earnings call UCT instead.

But I will say that using earnings call, it's really cutting our sample much shorter. Now, so here we wanna go deeper and gather more indication. On whether or index makes sense. So we wanna study the heterogeneity of this investment response across firms. In particular, accenti you would expect this investment response to be more concentrated among firms likely impacted by the tension.

And obviously large differences exist in the cross section of us firms. We're not aware of any metrics that would proxy for firm-level exposure to US-China attention. So we take on the task of constructing such metrics. What I'm showing you here is one metric that we construct to measure firm-level vulnerability to US-China tension.

The way we go about it is to internalize this efficient market perspective. Which is that if you think about firms likely impacted by the tension, you would expect these firms to have their stock prices, stock returns decline more sharply when tension rises, relative speaking all else equal. And so we go ahead and estimate the sensitivity of the firm's idiosyncratic stock returns to movements in our index, and we would interpret these firms with negative UCT beta, as we call this.

This is the sensitivity that we just estimated, UCT beta. We think of these firms with negative UCT betas, meaning that these firms whose stock returns tend to decline when UCT rises as firms being vulnerable to the tension. And we think about these firms with positive UCT beta. These firms having their stock returns, on average, tend to increase when UCT increases as being immune to or even providing a hedge against rising tension.

As a proof of concept, we will look at the industry distribution of UCT beta. So if you look at the firms with most negative UCT beta, they are concentrated in industries related to consumer electronics and equipment, and firms with the largest positive UCT betas, their concentrated industries related to insurance and utilities.

Now, what we do is to construct the indicator variables to flag those firms more likely impacted by the tension, relatively speaking, as indicated by firms having UCT beta under the 10th up to the forties fifties percentile of the UCT beta distribution. We put each of them one at a time in our investment regressions.

We also interact them with our UCT beta, sorry with our UCT news-based baseline index. So across board, we see indications that firms more likely impacted by the tension, as gauged from UCT beta, show a significantly stronger investment response. It seems that they're cutting back on investments more aggressively during periods of heightened tension.

Now, to the extent that trade linkage is prominent, we also look at firms operating industries that export and import most intensively with China. So using data from the US Census Bureau, we flag those industries, the top 10% exporting industries that those industries that export most heavily to China.

And we put that in our investment regression. We do see that exporting industries trade linkage does amplify the investment response. So very quickly, I'll show you one last set of results in the realm of investment analysis, which is that if we, as I alluded to earlier, US-China attention is really not restricted to trade conflict.

It covers really a broad range of issues. So to construct a measure that reflects that, we're thinking about these set of firms likely impacted by the tension. And we think that these firms would include those firms whose business success depends a lot on their market in China. And so these firms would likely to see their performance, and therefore their stock market performance, relatively speaking closely with the Chinese economy.

So what we do here is to estimate a co-movement of firms idiosyncratic stock returns with the Chinese stock market index. And we use indicator variables to flag those with extensive exposure to China as gauged from this co-movement metric. And putting them into the investment regressions. There are patterns emerging that suggest that firms that have has extensive exposure to China, as indicated by having this China exposure metric, as we call it, above the seventies, 80s, 90s percentile.

These firms are showing a stronger investment response to escalations in the tension, as gauge from our index. Now, just taking stock here, if you think about this average investment response, and you think about how that investment response varies systematically across firms, depending on firms UCT beta China exposure and trade linkage with China.

The confluence of results is in line with this idea that our index correlates with firm decisions that are indicative of tension. And if you take that and you think back on the chart, I showed you where our UCT index moves in lockstep with this average share of earnings call discussion devoted to the topic.

There is some data indication that our measure tracks the apprehension of business decision-makers over the issue. Now, what about policymakers? So as a motivating exercise, we're looking at the intensity of mentions of China in proximity to tension in the State of the Union presidential addresses. And that tracks our index fairly well.

Now, that's what they say. What about the politicians actions? So domestically, what we do is to study the congressional bills being considered and manually classified, flag those bills that can be characterized as anti-China. And so that's being plotted here in dark blue. And you see that there is a reasonably high correlation with our index.

And to look at their international, look at the international front. To the extent that we would expect rising tension to also manifest in the form of government disagreement here. What I'm showing you in black is actually the share of UN resolutions over which these two countries voted differently.

So every year, there are about 400 resolutions being put forth to your own council. We are simply computing a share of these resolutions over which the US and China voted differently. Now, with that many more exercises, I will simply point you to one set of exercises we do, which is that we also construct the Tension metrics for other country pairs such as US-Russia, US-Japan, US-Canada.

And the purpose of doing those indices is that it helps us make sure that some of the important characteristics of our index are unique to US-China. Not a artifact of the new space methodology itself, but we do utilize these other country pair indices as well. For example, I will share that the investment response I just showed you, it's something unique to US-China tension.

It's not there for the other tension metrics that we construct. Now, in the remaining couple of minutes, without showing you the exact econometrics exercise, and we can leave that for the discussion. I will say that we do move on and explore the economic transmission of US-China tension, with the caveat that causality causal inferences are inherently challenging to make in this context.

So in addition to the investment response that I showed you, we use firm level supply chain data. And we do find that over our sample escalations in US-China tension, increases in UCT are associated with supply chain diversification away from China. And we look at the US stock returns.

We find that UCT is being priced in the cross section of US stock returns in ways that are consistent with investor expectations of deteriorating economic opportunities during periods of heightened tension. There is a negative UCT premium there, and we also find negative transmission using VR analysis, skipping over those happy to discuss in detail in the discussion part of this webinar.

And the effects, the associations that I showed you, they predate 2018. It's not just driven by the trade war period. Before wrapping up, I will quickly mention that we did the decomposition of our UCT index into the first moment. The action component and the uncertainty component by separating our category t terms into those that likely indicate tension and those that do not.

Now, exploiting the independent variations in these two subindices, in these two components, we essentially feed each of them one at a time into all the econometric analysis we do. And what's consistently showing up is that they both have a significant effect, but the effects are always stronger through the uncertainty component.

So it seems that much of the adverse effects of US-China tension might be operating through the uncertainty channel. And this is some data patterns suggesting that the wait and see type of magnitude and the financial friction channel could be at work here. We also do daily version of the index, separating the index into subtopics.

Earnings call transcripts allow us to do it by industry, by region, and we also construct the counterpart index using chinese newspapers and also third party countr newspapers with an eye towards facilitating related research. To conclude, I will simply say that US-China tension does loom large. I think it's the multidimensional nature of it really calls for a holistic metric.

And we do that by applying machine learning and search based methodology on news textual data. And this data allows us to uncover some non-trivial cost of the tension economically. And such contractionary effects seem to be primarily operating through the uncertainty channels. And with that, I will stop and very much looking forward to learning your thoughts.

>> Steven J. Davis: Right, thanks so much, Bo. That was really fascinating. I'll start off with a comment and a couple comments. One is you can summarize a lot of what you said by the geopolitical atmospherics matter a lot for business outcomes and investment decisions. I just wanna note that the broad evolution of US-China trade is consistent with that conclusion.

So if you look at the growth of trade, say, imports into the United States from China, they were on a clear rising upward trajectory before China joined the WTO. But there's a very clear breakpoint in the data around that point where after China joined the WTO, you see a big increase in the trend growth of trade between the United States and China.

Now, what happened around when China joined the WTO, it's not that the US lowered its tariffs on Chinese imports. What happened is it took off the table, the recurring threat that had often manifested itself in proposed legislation in the US Congress. It took off the threat to revert to smoot Hawley tariff levels back the 1930s on China.

So once China joined the WTO, the US was prevented by its treaty obligations, as I understand it, from imposing smoot Hawley tariff levels on China because it then had most favored nation status. There's literature on this episode, but that's very much consistent with your emphasis on what you call uncertainty effects.

But as you also pointed out, it's really a mix of uncertainty and negative anticipation, first moment anticipation effects in there. So I think it's just the broad sweep of US-China trade experience is consistent with that theme. I'll make one other comment and then see what Aron has to say and go to the floor.

I like your firm level analysis. Your reliance on the firm level beta as a way to assess the exposure to US-China tensions is a very sensible one. But you can actually go further if you want to dig into the 10-K filings in the discussions of risk factors in the 10-K.

So I've done this in a paper with Stephen Hansen and Christian Seminary Ahmes. And what these risk factors discussions do is firms have fiduciary responsibilities to disclose their material risk factors that they know of and perceive. And so they will, at least since about 2006, I think, when these things really became more detailed, they will disclose with considerable granularity the nature of their risks.

Including if they're exposed to certain trade related or other, or geopolitical risks with respect to China, and that has material impacts on their business. It'll show up there, so I won't go into the details. Details of how we do that, but it basically gives you the opportunity to both more finely characterize firm level exposures to US-China tensions, but also to pinpoint it to the particular issues that are relevant to that firm.

And you may want to go down that path, so let me see what Erin has to say, and then we'll open it up.

>> Erin Baggott Carter: Fantastic, thank you, Steve, and thank you, Bo, for this really fascinating paper on an obviously very important topic. So I ask a question about temporality here, both in sort of like the nitty-gritty of how you analyze the data, but also broadly in terms of how we should interpret this piece.

So, in particular, I was wondering if you did use, or could use measures to look at projection language to capture whether these news reports are about recent events versus projections in the future. So, different types of reporting, and especially business reporting, will often suggest this might happen in the next week or the next month.

This is what we can look forward to versus a lot of more typical reporting is this just happened yesterday, and I'm curious about what type of coverage is most important for firm decisions, right? Recent events versus projections, so I was wondering if you have thoughts on that, but more broadly, right?

I like Steve's description of the atmospherics, everyone knows that US-China relationship has become much more tense over the recent years, both for long run economic change reasons and also sort of policy conflicts. But businesses should build that into their decisions, right? So I'm curious if you see less responsiveness to these indices over time as businesses decide that we live in an atmospheric environment in which there's more tension.

And maybe they move investments elsewhere or make more specific changes so they don't have to be quite as responsive to these measures or these risks, so I'm curious about how you think about that issue, thank you.

>> Bo Sun: Perfect, can I share some quick responses, Steve, before we open the floor?

>> Steven J. Davis: Yeah, please, go ahead.

>> Bo Sun: Okay, so I will say that, first, thanks to Steve and Erin, so much for the comments and suggestions, won 10-K, absolutely. It's on our list because, as Steve mentioned, that for 10-K and 10-Q filings, firms are legally bound to review their perspectives on the material risks to that.

And it could be a more reliable data source compared to earnings calls, where the conversations can be sort of swayed by the discussions in the moment, and so it's top on the list. And I will also mention to Erin's point that your first comment regarding the time horizon.

It's something that, from the way I understand it, our news based literature struggled with a little bit cuz it's always hard to pinpoint the exact time stamp on those news based discussions. But I will say that I think my understanding is that at this point we kind of reached a view that the monthly horizon would be something that we're comfortable attaching interpretation to the news based metrics.

Your comment is absolutely spot on, it's challenging to tease out, is this about the event happening tomorrow, or is it like months or even years down the road? We did at some point try with short run, long run looking for these terms indicating time horizons, it's a challenging task, but it's definitely worthwhile exploring.

I want to go back to Steve's and Erin second comments, because to me they share some similarity from where I said. So, what Steve mentioned was that there seems to be inconsistency, if you will, if you look at the upward secular rise in the bilateral tension, but at the same time, the trade and business ties have also been growing over time.

And what Erin mentioned was that there is this, like US-China tension just in the air, this is a broad atmosphere that everybody is processing at this point. I think both points speak to something, I think that's the unique usefulness of our index. Cuz if the goal here is to analyze what the precise economic facts of particular trade tariffs were the elimination of tariffs in the case of WTO entry, our index is not the perfect measure for that.

I think the literature has done this perfectly in terms of using the exact tariff changes or anticipated tariff changes. Now, what our index is used for, it speaks to this broad atmosphere that both Steve and Erin touched on, is that I think there is a sense in which, even outside the trade linked firms.

Even outside the directly impacted firms, there is a sense in which firms are broadly adjusting their forecast about their future barriers, their future constraints and their future economic opportunities in the face of rising tensions. And that's something that our new space, the metric, thanks to the methodology, thanks to the availability of data sources, would be able to say something, and this uncertainty channel points to that, right?

There is a sense that the economic effect could and have been showing up on a broad base beyond the directly impacted economy, now, with that, I'll stop and can't wait to hear more-

>> Steven J. Davis: On that point, it suggests it would be useful to break your index down into different aspects or areas of tension in the relationship.

And insofar as possible, separate off the trade related ones, which do lend themselves to other measurement approaches. But some of the important aspects of the relationship don't really lend themselves to other measurement approaches, so I think your index and your approach is especially valuable in those cases.

>> Bo Sun: Exactly, so we did, and we will post them online, where we did a desegregation of the index into topics separating our trade from, say, technology and national security.

What we haven't done yet is to look at the impulse responses to different type of tension, but that's definitely interesting, a very meaningful direction to go.

>> Steven J. Davis: Okay, so let's go to the floor, Tara, can we open the mic for Stephen Redding? He's got an interesting question that he wrote in during this session, and like to get that on the table.

>> Stephen Redding: Thanks very much, thanks for a fascinating talk, really interesting research on a very important topic. I was just wondering whether you could sort of further validate this approach by using previous geopolitical conflicts. So, for example, the confrontation between Britain and Germany in the early 20th century, you could imagine using your algorithm, applying it to historical newspapers, and seeing, do you sort of detect a pickup in tension in advance of World War I?

And what's also nice there is you actually have several events leading up to World War I, like the Agadir crisis in Morocco and so on. And then the war finally breaks out with the Balkan crisis in 1914, so you could even kind of say, can you detect which of these confrontation points is more likely to lend to war?

Does the Balkans look very different in terms of the newspaper textual search? That actually did let go to war, whereas the earlier Moroccan crisis actually doesn't look the same, even the attention picked up. So there could be lots of interesting aspects there, and probably those newspapers are available online, but super interesting.

>> Bo Sun: Thank you so much for bringing it up, Stephen. As I briefly mentioned, we do construct these counterpart indices for the other country pairs. I'll be mostly US centric at this point, but I will say that the direction that you point to, which speaks to this cross-country. And also tracing back in history to increase our understanding about what triggers war, what kind of tension, and then also in terms of broadly speaking, utilizing cross-country and longer time series dimension to look deeper into the economic effects and drivers of the tension also.

So that's something that we're seriously thinking about doing. So thank you so much for bringing it up. I'm glad that we did.

>> Steven J. Davis: Yeah, I'm with Stephen, and there's just tremendous scope for the intersection of historical and economic analysis from these approaches, and I'd love to see more of it.

Maybe Steven's gonna go down this path. So it'd be great to see, too. Let's turn the mic over to Elizabeth Elder.

>> Elizabeth Elder: Great. Thank you so much. I was just curious a little bit about time ordering here. I'm always curious the extent to which news reports lead or follow changes in elite perceptions of things like this.

So are they reporting on concerns that business leaders already had, or are they creating concerns in the business community that didn't exist already? So I'm just curious, if you look at a plot, not just of changes in investment post changes in news coverage, but changes in investment pre changes in news coverage, when exactly does that pattern start?

Is there any anticipation here? I think for business, but also especially for politics.

>> Bo Sun: Perfect. Yeah, I think that the direction of arrow is definitely two ways, but I think we could do more in terms of the Li lag relationship and teasing it out. Right now, our analysis is just making contemporary correlation statements.

But I agree that it will be interesting to see. Is it more that it gets discussed in the news and therefore firms managers start to talk about it, or it's something that going from the business information flow, we would look deeper, more carefully into that. Thank you.

>> Steven J. Davis: Yeah, I'm gonna read a question here first from Brett Carter, who I think is in an airport.

So that's why I'm gonna read his question. He's asked the following. To what extent could your measure be picking up tension with other countries that have generally similar interests to the US, such as Japan? He also says, really enjoyed the talk and delighted to know you read the propaganda book.

I think he's referring to his book with Aaron on propaganda in autocracies there. So that's an interesting book for those who are listening. But anyway, he's asking about it's a question about whether you're really isolating what you want to isolate here.

>> Bo Sun: Absolutely. I think that it's very insightful, it's fundamental question and that's why we did this large number of evaluation exercises that I didn't exactly bore you with in its entirety at least.

So one thing I will mention among the things that we did that would speak to some extent to brand's concern is that we did a large number of permutations regarding exclusion terms. That's where we will restrict the UCT articles to be those that do not mention Japan, do not mention the Middle east, do not mention the other regions.

And in that exclusion terminal type of exercise, we also exclude mentions of sports, olympics, so on and so forth, not just the war type of issues. Now I will say that if you look at one of the gazillions of appendices, we have in the paper, that these results, these alternative indices with exclusion terms show such high correlations like above 0.95, that we're not so concerned about the false positives there.

And I also wanna share one more thing on that in that regard, which is that we also have an alternative index where instead of doing what Steven Scott did in their EPU paper, doing the search based methodology. We actually look at more of the intensive margin, so to speak, where we download all the newspaper articles and for each newspaper articles we assign a UCT score measured by the frequency of occurrence of tension words in close proximity to China.

And so that intensive margin type of index also shows very high correlation with our baseline index. So I think that to some extent also addresses our concern about this type of false positives. And last thing I will say is that I think what gives me most confidence in the informativeness of our index is that it seems to be correlating remarkably highly with the views expressed by firm side and policymakers, both in rhetoric and in action.

I think to the extent that when we quantify the apprehensions of the other, not just not newspaper readers directly. But when we quantify the perception of the firm side and policymakers were essentially using independent, different data sources and different approaches, to the extent that there is a lot of common variability.

I think there is indication that there is signal value in our index. But great questions, great comments. That's something that we thought a lot about.

>> Steven J. Davis: Okay, we've got a Liz economy had to drop off, but she had some interesting questions. She says fascinating talk. And so I'll just go through one or two of her questions.

She says, for your business leader perceptions, do you also use or refer to the surveys from Amcham China and the US-China Business Council which ask questions that are directly relevant to your study? And she says their findings are in line with your results.

>> Bo Sun: Absolutely. I last minute took out that slide on that survey.

>> Steven J. Davis: Okay.

>> Bo Sun: And if you look at that survey, you'll see that the rising tension just routinely climbed to the top of the list of challenges that firms face. I completely agree. That's in line with what we do. Maybe next time I present it, I'll put it right back in.

>> Steven J. Davis: Okay, which survey are you talking about exactly here?

>> Bo Sun: Let me share-

>> Steven J. Davis: I'm curious about it. We can put it up if you want, if you have a slide.

>> Bo Sun: So I think this is the survey that's being discussed here. If you just look at the recent years, it's definitely top of mind.

>> Steven J. Davis: So this is a survey of business leaders in China of American firms. Who gets this survey?

>> Bo Sun: So I think that this is mostly the, I can be wrong about this, but it definitely includes American firms operating in China. And I think, I believe that there are also.

Chinese firms exclusively operating in China as well.

>> Steven J. Davis: Okay, and Liz also asks, have you considered looking at how bilateral tensions affect Chinese business decisions in the United States and I guess related to the United States. So, cuz you gonna also try to focus on that rather than the focus exclusively on American firms.

>> Bo Sun: Yeah perfect, so what we are doing right now is to do the firm level analysis, then doing the counterpart of all the econometrics using the Chinese firm level data. And there so far what we did was that in response to the US newspaper UCT index, it seems like what's surprising is that Chinese firms are also showing negative responses to increases in the tension as measured using US newspapers.

And we also are looking into the Chinese firms investment response to Chinese newspaper based UCT, there we primarily focus on people's daily as the flagship news outlet. And we also find some responses there too, so we should include these results in the next version of our paper too.

>> Steven J. Davis: Yeah on that point, I'm just curious whether you get more or less explanatory power from Chinese sources or US sources in explaining the outcomes for both US firms and Chinese firms. So if you're a Chinese firm and you're thinking about your investment decisions in the United States, what is more influential in your decision making?

It's not obvious what the answer would be.

>> Bo Sun: I don't have an informed answer right now we'll look into that.

>> Steven J. Davis: Okay.

>> Bo Sun: I also-

>> Steven J. Davis: Go ahead, go ahead, please go ahead.

>> Bo Sun: I just really quickly jump in, I think that if you're, it would be great to see the comparison with the responsiveness of the Chinese new sources.

I think in this particular application, instead of using the people's daily, it might be nice to use the economic press like the economic observer, or Caixin or tai Ching, just because those are traditionally have more of a scope to report honestly, particularly on financial and economic events. So you might find more useful sort of signal in that data.

So we are looking for ways to expand the set of newspapers we use in China. But I will say that one thing I believe, which echoes Erin, what you find in your fascinating book is that, I think the more commercialized the Chinese newspapers tend to go with the people's dailies line on issues that count and especially on China stance regarding the relations.

So I would expect a lot of co movement there, but it's always interesting to check and explore.

>> Steven J. Davis: Okay, very good, this has been a fascinating discussion, really interesting work, and obviously there's more to do here, but you've already covered a lot of really interesting ground and thanks for joining us.

Thanks for this, and we'll sign off for the recorded session now, but we're going to stick around for an informal discussion for anybody who wants to participate. Thanks so much Bo, that was great.

>> Bo Sun: Thank you so much.