December 13, 2023
Hoover Institution | Stanford University
Our 21st meeting features a conversation with Thiemo Fetzer on How Big Is the Media Multiplier? Evidence from Dyadic News Data? on Wednesday, December 13, 2023 from 9:00AM – 10:30AM PT.
Thiemo Fetzer speaking on How Big Is the Media Multiplier? Evidence from Dyadic News Data?
The Hoover Institution hosts a seminar series on Using Text as Data in Policy Analysis, co-organized by Steven J. Davis and Justin Grimmer. These seminars will feature applications of natural language processing, structured human readings, and machine learning methods to text as data to examine policy issues in economics, history, national security, political science, and other fields.
Thiemo Fetzer is a Professor in the Economics department at the University of Warwick in the UK and at the University of Bonn in Germany. He is also a visiting Professor at the Grantham Institute at the London School of Economics, a Visiting Fellow at the London School of Economics and a Fellow at the National Institute for Social and Economic Research. He holds further affiliations with the Centre for Economic Policy Research (CEPR), the Spatial Economics Research Centre (SERC), CESifo, and the Pearson Institute at University of Chicago. He serves as a Theme Leader at the Centre for Competitive Advantage in the Global Economy (CAGE) at University of Warwick.
He has published extensively in leading economics journals such as the American Economic Review, the Review of Economics and Statistics, the Economic Journal and the Journal of the European Economics Association. Thiemo’s research has been featured in the New York Times, the Washington Post, the Guardian, Foreign Policy, Le Monde, and the Financial Times. He has served as consultant and advisor to a range of national and multinational organisations. He won a European Research Council Starting Grant for his interdisciplinary research project MEGEO – Media, Economics and Geopolitics and was awarded the 2022 Phillip Leverhulme Prize in Economics.
Steven J. Davis is the Thomas W. and Susan B. Ford Senior Fellow at the Hoover Institution and Senior Fellow at the Stanford Institute for Economic Policy Research. He studies business dynamics, labor markets, and public policy. He advises the U.S. Congressional Budget Office and the Federal Reserve Bank of Atlanta, co-organizes the Asian Monetary Policy Forum and is co-creator of the Economic Policy Uncertainty Indices, the Survey of Business Uncertainty, and the Survey of Working Arrangements and Attitudes. Davis hosts “Economics, Applied,” a podcast series sponsored by the Hoover Institution.
Justin Grimmer is a senior fellow at the Hoover Institution and a professor in the Department of Political Science at Stanford University. His current research focuses on American political institutions, elections, and developing new machine-learning methods for the study of politics.
Welcome everyone to the Hoover institution workshop on using text as data and policy analysis My Name is stevenh Davis uh Justin Grimmer and I select speakers uh and moderate the workshop Tara Mahan is our Master engineer and Cecilia Chen Keeps Us organized today’s guest is Teemo Fetzer of Warwick University he will present
His paper t titled how big is the media multiplier evidence from dietic news data it’s a paper co-authored by Timothy Beesley and Hest Mueller here’s our format Teemo will take about 30 minutes to present then we’ll turn to a discussion if you have a question or
Comment put it into the Q&A box and depending on the flow of these comments and questions we may paraphrase them combine them or ask you to State them directly we’ll run for about 60 minutes then we’ll turn the recording off and we’ll have another more informal session
For anyone who wants to stick around uh that’s Offline that that won’t go into the videotape session okay uh with that Teemo the floor is yours thank you so much uh for having me this is it’s a great pleasure to present this work um this work actually started in 2014 so
It’s been a while um it was just recently accepted for publication um so this is Joint with Tim and hes what we do in this paper is essentially trying to quantify the extent to which um selective media coverage can distort um the uh economic effect potentially or exacer ex asate potentially risk
Perceptions they’re associated with relatively rare events one of the motivating figures that we have for this work is uh from our world in data where you know if we look at what the media is reporting in on Vis of what people are in this case dying off there’s a vast
Difference in uh in in reporting skew so if we look at the number of news articles that are in the guardian or the New York Times that cover for example uh deaths that are related due to terrorism or uh or homicides they take up a disproportionate amount of media real
Estate so to say um whereas actually in terms of the actual causes of death the side Killers like cancer and heart disease obviously um um you know matter materially much more um we studi this phenomena in the context of uh uh you know obviously this is relevant for
Economic decision making because um uh things factors like heart disease have a much higher human death toll um but uh you know humans seem to be drawn to kind of gory images if it bleeds it lead um and uh uh nowadays where media has moved more into being in the marketplace of
Attention rather than in the marketplace for information um and in this paper we study essentially the Distortion that this can induce essentially amplifying economic Cycles or potentially economic shocks um so I don’t want to motivate this much further I think I’ve done a lot of motivation in this uh one slide
Already um what we do in the paper is to model the uh economic effects of negative new shock um and we study the extent to which there Amplified by media coverage and through that uh we try to quantify the uh what we call the media multiplier as sort of a approach that is
Grounded through a a a formal model uh a data generating process that we posit um and that we then sort of bring to bear on data that allows us to both address uh a causal identification challenges as well as also provide us with a setting
Where I think we can uh uh you know do quite decent quantification um so let me just motivate the uh story a little bit or the specific context um this is a type of violent event that we’re studying in the paper which is um the sus attack in
Tunisia that took place in 2015 uh when about uh several gunmen you know started shooting up tourists that were just visiting a beach um and that has resulted in 38 fatalities um in this specific case the vast majority of victims in this particular instance were British Nationals and this uh event dominated
The new cycle for about a week and even then resulted in the um the this the government the UK government to repatriate its tourists while uh you know German Nationals were not evacuated so you know just imagine this is a very strong and pronounced shock and had significant implications for policy
Making and obviously also had economic implications we can study this through a granular uh card data payment card data that it took us I think three years to secure actually Four it started in 2014 um um um which basically provides us a measure of service sector trade which is
What tourism is tourism as a form of service sector trade um and here what I’m plotting is the activity the card activity that’s associated with British issued British Bank issued uh uh Master cards it’s all aggregated you know and of non non-disclosed data that we’re working with here um British cards
Issued uh that are used in Tunisia over time and German cards that are used in Tunisia over time and what you see is around the month it’s monthly data we see that there’s a significant kind of decline relative to after having residualized moving removing some you know lever shifters as well as Trends
From the data we see that there’s a significant decline in the uh uh sort of amount of spending that can be attributed to British issued cards in Tunisia and that effect has been much more pronounced relative to uh German issued cards in Tunisia ultimately that is sort of the differential that we’re
Trying to explain and to what extent this type of differential of course this is a very specific context a very specific event that was very you know high intensity um but essentially what we’re trying to do is explain to what extent this differential response can be attributed to the different intensity of
Media coverage this is not the archetype event that we see in our data but it’s sort of one where it’s just very plain obvious In The Raw data to visualize and motivate the uh exercise that we’re doing in this paper um so we’re exploiting diic data in two ways it’s
Diic data that because we have network data both from the side of the card uh payment cards that are being used so we see that uh in five Des tourist prominent tourist destinations um we’ve gotten data uh on card activity both the number of active cards the number of
Transactions as well as the spend um that arises from about 128 card issuing countries so it’s granular high frequency monthly service sector tourism uh data um covering the time period from 2010 to 2016 we got the data in I think October 2017 after searching for three
Or four years of finding a partner that was willing to share that with us um the second and that’s of also novel novelty in in this sort of type of work is that we’ve built a dietic news Corpus uh data set so we’re capturing um um data on how
Other reported on each other’s news um which ultimately uh you know is what we want to want to study here which is how for example a violent event that is taking place in Tunisia is affecting the media reporting or the media representation of Tunisia in a set of potential tourist origin countries so
Think of this as like a directed uh directed graph um um um but of course since the five destinations are predominantly service sector uh uh exporters meaning they import tourists they they have tourists arriving at them um the focus is the uh service sector trade from the perspective of you know
Mostly uh you know Western European and so on uh uh countries into uh into these uh destinations I think when we started this I think we’re probably one of the first to build such a dietic data set on the news data side as well as on the
Card side and combine that together um in in in in that in that paper um let me just talk a little bit more about the data description and diagnostic um how we sort of Leverage the Corpus how we try to filter um the uh news articles that are diet specific and that are
Capturing violence uh using I mean I think nowadays probably pretty redundant NLP techniques um and uh um how we then take that to bear on the data um Building A reduced form exercise but also then building a what we call like a data generating process representation um that we then let bear
On the data um so we’ve obtained This Confidential aggregated data from Master Card capturing these three things uh from 2010 to 2016 it’s a monthly data set so they wouldn’t part with more granular data uh for confidentiality reasons or trust reasons or whatever uh at the time this was fairly new for for
Them to actually build these type of Partnerships and I think we’re probably one of the first ones that actually uh were benefit benefiting from such a data donation the first time I actually it’s a nice anecdote when I spoke with some representatives of uh mastercard’s main competitor um they were actually quoting
I think a single time series of the date that we’re working with Like the quote was about $10,000 and I said well this is some business you’re in but maybe you get your business wrong because you should be probably in the type of uh you know Consulting business rather than so the
Data merchant business um and I think some of the firms in the sector are now picking up on this quite quite significantly but still do other players uh that are entering that realm um so we construct a diet level panel data set we’re focusing we have data for 140
Credit card issuing countries it’s credit and debit card um we focus on a subset um uh um where we have a balanced panel uh um you know just to make things a bit more more more you know it’s just a bit more elegant but it doesn’t actually change anything fundamentally
The second data source that we Leverage is was also quite painful to collect because you know we don’t have apis and and whatnot we didn’t have API access uh was content data from Lexus Nexus and faiva um that essentially through which we try to identify the population of all
Destination specific news coverage so uh uh what we did is for every tourist originating country we’ve I tried to identify a large consistent source that is available over the whole time period actually with one year lack as well um that we query essentially uh um um in terms of whether
The country name or prominent City location names appear in that in that article that would define the population of articles that are capturing a destination so it’s Tunisia Morocco uh Egypt uh Israel and Turkey uh that we’re working with um and this uh essentially sort of provides us like
The Corpus of all the articles that are covering set destination um we actually literally had to download these piece by piece we had some Ras like that were downloading the data and batches of 100 which was a very very unhappy job because at the time the
Uh well we just couldn’t afford or it was not possible to get API access to either Lexus Nexus or factiva in total we have about half a million articles across 20 different languages that we then for convenience translate to English um so what we need to do uh naturally is identify among
This Corpus um the articles that are likely indicative or capturing uh violent events uh um that we think are you know what drives the underlying mechanism of how people who might selectively read uh news uh uh might respond to uh in terms of their consumption or travel decisions um just
To give you a sense uh of what this looks like and there’s usually always some feedback on this which countries are being covered and what is the source origin origin um um is it appropriate is the New York Times an appropriate representation of a newspaper in in in
The US when I originally designed this this this paper and and as there a dual paper of this where we look at reporting on NATO casualties in Afghanistan the idea was to find an a both a left leaning a cist and a right leaning uh newspaper for especially the Democratic
Countries uh that were true contributing Nations um it is simply for the content aggregators that we had access to it was hard enough to identify a consistent source that goes back to 2009 over the time period but that was originally the idea uh um to look at the different way
That you know the same story might be slanted or spun uh across different types of sources a usual question that would come up um we also flag up that in some instance the only source that we had consistently available are news agencies so for China for example you
See sinoa and that obviously is not a representative Source but all the results that we uh work with are robust to the exclusion of uh news agencies or agency sources because of course there’s two factors that matter whether something gets reported on and whether that something actually is appearing
Let’s say in the daily new diet that the average household or consumer might be subjected to um yes so this is sort of just a comment here on on on the sources in terms of the countries that we cover the dark gray countries are all the countries for
Which we have both data on the underlying sources so we have both a news Corpus as well as data so we basically cover pretty much the on the substantive uh terms uh much of the world economy uh and all of the G20 countries so um it’s definitely captures most of the countries that
Would be service sector importers uh via sending out tourists um in terms of the machine learning I mean I think this is I mean now with llms and so on and so forth this is actually pretty uh pretty simple uh um the what we what we do is um we um
Use uh you know human coders to classify a stratified random subset of the Articles this half million articles and where the stratification was done because obviously violent events thankfully are sufficiently rare but it also causes a problem because obviously any classifier would have an tendency to just classify all articles as being
Nonviolent or capturing not reporting on nonviolence if 95% or 96 98% of the articles are covering nonviolence so what we’re doing is um we use some conflict event data sets a broad menu of conflict event data sets that are ultimately we consider to be the superet
Of all the let’s say uh news uh um about the actual factual events we take this as an anchor and then over sample articles that uh around like you know a onewe window or three three-day window around around an event so that we address the inherent class and balance
Problem through that through the training data uh uh selection um so we use this supervised uh machine learning uh um where we classify articles in two ways whether a text is indicative of whether there was any violence or whether text is indicative that tourists were the target of violence these are not
Nested uh uh uh um as but but you know you can think of them as being uh super sets uh um but in terms of the tourist targeting obviously this is a much more narrow much more specific type of uh article um we U use U um a very simple
Naive base classifier and two sets of random Forest classifiers um that we sort of for this a little bit iteratively uh where we sort of recode uh post the first uh uh you know sort of run of it uh um to identify essentially uh to see how many
Sort of Articles we might miss or the classifier has a low performance uh in some area domains since the analysis is all done with a monthly diic uh card spending data we have to essentially uh uh turn that daily data set or article level data set into monthly monthly uh
Aggregate uh um and our preferred measure of uh violent news reporting is uh expressed as a share where you know and we talk about this in the modeling section the numerator is the article the denominator is the is the number of Articles where we have to add plus one
In case uh a country never appears in the news of let’s say Tunisia never appears in the German news which has very interesting nonlinearities uh and and features that we try to represent in the model um since it’s a short presentation uh I can sort of you
Know just quickly rush through this I think I’ve said most of this um you know we use an ensemble uh uh um agreement uh um to uh classify an individual article aggregating so the linear modeling that we get from the naive phase with the nonlinearities to
Try to sort of combine so of the uh the different features uh where you know random Forest have a tendency to overfit whereas naive Bas is is is is linear we’re trying to S trade off the two in a very agnostic way but by giving each of
Them a third or a weight we could have optimized this as well um we’re not using the base optimal cut off um but we’re using sort of uh something that because we since we aggregate the data the aggregation helps get rid of some noise uh um um the in in the process uh
We work with slightly different cutos but all the results are robust uh to using different cuts um just to look at the class in balance out of the 450,000 articles 16,000 are covering or 177,000 are covering anything related to violence and about a th thousand articles are covering a violence with a
Specific mention of tourists being targeted um this is the just the headlines of some example articles with uh tourists being uh with sort of the general violence topic and there we see we’re picking up quite a bit of uh let’s say kinetic force or milit more sort of
Military type engage engagements uh um so like clashes of pkk rebels uh with the Turkish military um or uh sort of violence on the Egyptian scai uh um which might not directly you know be relevant for tourists but it’s obviously a country specific risk uh um that
Matters um this is um the uh other outcome or the other classification or the other label so to say which is tourists being the target of violence and again if we look at the uh articles um they seem to I mean the classifier seems to perform uh uh reasonably uh
Reasonably well um let me now just walk quickly through the reduced form evidence and then the model and the model bit um there’s a whole lot of exercise that just try to situate this work within the existing literature on on on this uh um where uh uh we s look
At violence data violence separately from news is there content in the news reporting that seems to explain variation above and beyond what sort of a simple violence level controls would do um which is what the first generation uh of papers in the early 2000s did on like regressing violence on some
Economic outcome um so we have this sort of news measure which captures the share of news on a specific origin Like Home Country destination country pair we lack this by one month in this very simple reduced form uh setting and uh we have a whole exercise justifying the use of
That share measure which is of backed up by again our of the modeling exercise that we do um so if we just look at the reduced form what this is saying is that well if you move uh if a country pair moves and it’s reporting from zero to
100% you see a near uh 80% decline in tourist activity um and that is quite robust to you know multiple uh you know multiple different ways of accounting for linearities or nonlinearities different time Trends origin destination specific uh linear uh nonlinear time Trends and a whole lot of other
Exercises that uh you know are now sitting in various appendices never to be yes just a clarify question the an increase in violent news share to 100% means all articles um about that destination country and that origin country or or about violence yes okay so that but
That’s a huge shift what’s what would be something more within the or maybe that is within so that’s exactly part of the story which we which I’ll try to highlight in the modeling I’ll get back to this if I don’t uh please remind okay um we also do some I that exploits
Because of course you might be concerned that uh you know uh you know tourists might terrorists might Target specific tourist nationalities um which you know there is some evidence on tourists that terrorist targeting so the timing might not be exogenous I think um um what we do here
Is we exploit variation in the casualty distribution the nationalities of casualties for some events where we were able to identify the casualties the the country of origin of the casualties to then look at whether we can document this treatment effect through spillovers so the idea is very simple we’re saying
Is Well what is the effect of a German casualty in Tunisia on the spending of austrians in Tunisia and we ignore the full direct treated diets which of course highlights that we likely have a violation of the Suva assumption because there’s information spill over us because the German um Char is more
Likely to be reported on in the Austrian news because we share a common language um so this is just a cute extension which s of highlights I mean of course it’s not a superpowered instrument but we can identify um these treatment effects through these information spillovers which address a whole lot of endogeneity
Concerns um there’s a dynamic to this um but let me maybe jump straight to the modeling uh section um because the reduced form is cool but also a little bit limiting and for the model what I like to do is uh present this as sort of
Like you know we had this idea of like can we represent this in something that looks a little bit like plate notation uh um and so this is like a a very rough attempt of of lat notation we posit that there is essentially countries there’s a latent State that’s unobservable a
Country can be safe or dangerous that’s sort of evolving according to like a Markov chain process uh um and that Markov chain is you know governed by some transition probabilities uh and persistent parameter persistent par persistence parameters um what is observed aable is the actual violence time series in a specific destination
And the violence process is governed by this latent State and then you know if the country is in the violent State you draw from you know violence realizations in this case we assume the simple normal distribution with a date specific normal distribution with mean and variance the
Other bit that’s observable is the bad news and all the other news uh um so that’s in the bottom hand the two notes slightly gray shaded so we have bad news that it’s origin destination specific and all other news which again um is governed the evolution of that is
Governed by the latent uh State uh uh Evolution now we posit that there’s two types of consumers um one of them I call we call and there’s no normative Notions to these labels there’s the sophisticated and non-sophisticated consumer the sophisticated consumers are the ones that essentially consume the super
Superet of all the news which should cover just the raw facts uh so think of this as like the super set of all potential articles which means the you know it’s our best estimate of what the ground proof violence data is in a specific uh destination so this is this
Uh Pi capital capital Pi uh DT the other set of consumers call them naive again there’s no normative notion here to this labels um but they predominantly consume news that’s available to them through the media reporting so they form their beliefs about the latent State and the
Evolution of the latent State based on consuming uh media reporting uh um and of course bad news and all other news this is essentially the inputs to our model um um that we then bring to bear on on the data we incorporate a lag structure uh um and
Obviously there’s we we allow there to be different weights of both late both um both uh let’s say naive versus sophisticated tourists or potential travelers to be in there the lack structure we embed to incorporate um the uh you know the delays or the forward-looking nature of some planning uh travel
Decisions um I’ll because yeah I have about three minutes so I’m not going to if that’s okay not bore you bore you with this in terms of the form formalization uh um um we uh we our our sophisticated first as I said they observe the objective violence data when
Forming beliefs about the underlying State whereas the naive tourists mostly rely on the news uh reporting um the evolution of the violent state is governed by um you know a mark of process um that we then sort of bring to bear on the data um and this is sort of
The output of that Mark of switching model so is the probability of the state being dangerous that’s just imputed from the extractor from the from the violence uh country level violence time series data uh so to say so you see the switching and and and and non switching
And for the naive tourists um we model the arrival of bad news and all other news as being governed by a negative binomial distribution that’s conditional on the latent State uh where we calibrate the arrival rate of uh news bad news based on just the overall share
Of uh uh uh uh the probability of news uh uh being of an article being violent basically um through this it we can sort of you know you know apply Bas rule to study the evolution of both the news news based uh belief as well as the uh sort
Of sophisticated uh beliefs and get essentially some of the nonlinearities that the maros switching model implies into into into the model that the reduced form modeling cannot directly incorporate um this is a distribution of the beliefs uh um for naive tourists um about a country being uh safe versus
Dangerous uh um uh again because we do this for both you know latent States uh when the when the place is dangerous versus a place is is safe uh um and uh we then can estimate the medium multiply is essentially down to the mixture of consumers uh in a in a representative
Consumers that form these beliefs based on uh um based on uh naive versus sophisticated uh belief this is sort of our conceptualization essentially of the media multiplier what is the share of sophisticated versus nons sophisticated agents uh in that economy which we calibrate uh based on a a grid search so
We’re trying to kind of optimally fit essentially the parameter distribution we do an exercise to just show that they signal the statistical processing of the underlying data induces some signal and nonlinearities which we uh which sort of uh you know are not sort of you know there is the statistical processing of
The data refines the data the raw data in a way that preserves a signal um I just want to finish off based on uh sort of some calibration of what happens to a new shock um and this is where uh you know the point that you made Steve metas
Um the medium multiplier the medium multipliers size is a function of the numerator as well as the denominator so if you are Tunisia and your country is hardly ever in the news one bad news you know induces of course you to go bang bang on the reporting but because of the inertia and
The lack of data of other reporting that allows consumers to update their belief uh um it results in a degree of stickiness that makes the whole envelope of sort of the recovery out of that new shock much more costly um and and so the other news arrival matters which of
Course is something that we could uh use very much to understand lots and lots of phenomena that are happening around the world U that I’ve been sort of yeah somewhat uh working on um so I’ll just leave out here we can calibrate this uh also we do some speculation out of
Sample validation uh um around the size of the medium multiplier across different contexts using different corpora um but I’ll just leave it here um thank you so much uh for for listening to me great thank you teimo um that was really interesting I want to ask you I’m I’m
Struggling to just interpret exactly what the multiplier means conceptually and maybe you can help me here the thing I’m one thing I’m struggling with is the ground level truth that the sophisticated uh um consumers know in your in your setting is I understand that that’s something that’s constructed from
Multiple sources and I think largely after the fact so I’m not sure how anyone would know that in real time so that’s the thing that’s being multiplied upon but your your your thought experiment is that there’s some way you know these things in real time that would inform the tourism spending
Decisions and travel decisions of the sophisticated agents the other agents are just responding to what they see in the news but you can see my struggle here I don’t really know what the what the the base that’s being multiplied on is it’s really a conceptual object not a
As I understand it not a it’s an as if object not a not a real time thing you could look at like let let me just make an analogy if we were doing infectious diseases and the risk of the country specific risks of infectious diseases then a sophisticated traveler may just
Go go to the World Health Organization site and see what the latest advisories are so you’d have a very clear concept of what the ground level truth was in real time that seems missing in this context or have I misunderstood so so I think uh again the
We just posit that there’s these two types of Agents right and you can very much think of the sophisticated agent as the one that actually looks at advice um which is uh uh you know however the travel advice is obviously incredibly potentially incredibly sticky so um the
Uh the uh you know travel advice hardly ever gets updated or it gets updated with uh you know significant uh lag um even though there might be no material risk right because the decision to uh adjust travel advice is also something that can be politicized and is actually
Used uh in uh bilateral kind of U right debates right but your your your version of ground level truth in this application is not travel advisories as I understand it it’s some it’s The History of Violence it’s The History of Violence were which were which constructed after the fact and I’m
Guessing that often constructed from news sources but I don’t know that yes yes exactly it’s constructed from new sources but again we look at this that that these new sources are not country specific so uh um so you know whether a tourist got killed you know in Tunisia
Is not specific to each of the countries that might be sending tourists there so it’s it’s sort of like you know it’s an objective risk of becoming the victim of violence irrespective of nationality so to say I mean got it that’s a good way to put it but again correct me wrong
It’s an objective measure of risk but not not not a measure that was available in real time it’s one that’s been constructed after the fact that’s what I that’s what I understand that is that is true that is true uh uh but that is sort
Of like any you know we can make of course an out of sample prediction based on you know the you know based on that what the what the probability is that a place would be be dangerous uh at any given point in time because there’s persistence in these in the Marco
Switching model right it’s not like the places go bang bang all the time because there’s baked and persistence and so at some level it is we we think of this as a latent indicator of risk that I think has some signal we try to Crown proof this with uh country travel advice data
But getting Country travel advice data dynamically over a long period of time yeah we didn’t want to go down that route after yeah I understand I understand that it’s just just one last comment the the conceptual model that you’ve set forth that you use to estimate the media multiplier It just
Strikes me as more apt for the information environment that surrounds infectious diseases than violence for the reason that there is a Fairly reliable as I understand maybe I’m naive A Fairly reliable source you can go to to get travel adice advisories with with respect to infectious diseases in in near in near real
Time exactly no I I’m I’m totally with you and it’s also one example that we highlight in the introduction that you know it mimics very much that setting uh there was just no infectious disease environment that we could exploit over that time period in these five destinations that you know uh would give
Us because obviously there’s a causal identification part as well right that that you know we need to uh we need to address okay thank thank you let let me let let Justin jump in here I suspect he’s got some comments yeah so this is great I had a lot of fun uh thinking
About this uh project I think there’s a lot of facets to it that are are deeply fascinating I have uh sort of like two broad uh sets of questions one is sort of about the it sort of Builds on what Steve was talking about the information ofi environment available to to
Travelers and so part of that is the news part of that um that I was trying to puzzle through a a little bit is the travel advisories or or travel restrictions that sometimes one country will place on traveling to another that may not be based on a sort of
Sophisticated model that also could be based on reactions to news coverage or um sort of latent concerns that a country has about sort of risks but also that information environment could include positive statements from a country that’s sort of advertising itself as safe or you could imagine safe
In particular ways despite uh the certain kinds of things you may have heard um yeah and and that can interact I I think in some pretty interesting ways with when start thinking about the consumer model even if we just think that they’re making the sort of dietic
Decision you know do I go to Tunisia or not uh lots of other things are going on other than the news coverage yes uh it’s one of the exercises that we’ve now had to put into the append IND because you have to fit it to 45 pages but we looked at press
Freedom um as one of the features uh um the extent to which a country has a free press when reporting on violence and you see that there’s interesting heterogeneity let’s say Russian tourists traveling to Egypt do not respond in the way that European tourists would travel uh um because uh well reporting on
Violence in Egypt uh is suppressed in in Russia and there might be you know geopolitical alignment or incentives to keep that out of the news in the same way and so there is some interesting heterogen that is playing around with the consumer side but also with obviously the producer side right as in
Like the tourist destinations might have incentive to suppress a certain negative uh news coverage the same way this mechanism or the the model here you know again I look at this very much from the perspective of projecting soft power or constructing like a soft power index which is something something that I want
To do in my ERC if I ever get to it uh um because um obviously uh um you know there is two things going on any type of news coverage might help you see that this is part of country’s uh Sovereign strategies to uh uh I mean that often is
Gets labeled as greenwashing or uh or potentially Sports washing and whatnot um to sort of improve the external image that the countries uh have um and which you might be legit legate not legit legitimate I you know that this is not not for me to judge but or also bringing
In sports influences into into their respective economies the effect is not on the own economy the effect is on the origin uh you know where the respective David Beckham are for uh coming from you know that’s the audience that are being played uh uh uh when when we when we
Think about this phenomena um and so again I I as as Steve as you highlighted there very much applies to the context to a pandemic you know where we have this sort of objective source is much clearer right uh um but there’s lots of other factors again that you know in the
In the broader context the probability of dying in a car accident in Egypt is much higher than the probability of dying from a uh from a from a violent event and there is this tendency of course of the media to uh potentially skew uh information in a way that
Selects and attracts people’s attention um that can be both weaponized but also is something that can lead to over and underreaction producing volatility um that you know might adversely affect development outcomes and I would very much like to say that this also has I think relevance for foreign direct investment decisions uh
Or in general investment decisions especially of retail investors which in the context of uh tackling the climate crisis is particularly important because we need to facilitate Capital flows from rich countries to poor countries and if that involves mobilizing retail investors um and of course if there is
Such negative news bias and a history of bad news I mean I always gave the example of um you know I I was born in 86 uh so I grew up with images of starving East African children um you know that just sticks that type of is a cohort almost a generational cohort
Effect so if somebody says Ethiopia to me that’s the first image that comes into my mind uh when in fact it might be very devoid from the economic reality at the time but it might deprive that region of uh uh economic activity or Capital flows that could help it you
Know develop its domestic economy okay I have a I have a narrow question and then a bigger question so I ask the narrow one first and then I’ll get to the bigger one um the narrow one is you know violence is sort of a standin I think for unpleasant experience while abroad
And um you the next sort of proximate unpleasant experience I was trying to think about was a bad interaction with the government in the other place and like the most Salient example of that is you occasionally hear these stories of you know someone uh from Australia goes
To Thailand and they brought in drugs or something like that and then they end up you know in prison for 30 years or something like that any any sense of how prevalent those sorts of stories are do they they work in similar ways different ways is
Uh do do people react to those I would would love to study it we only had four countries and I think they the risk of uh I mean it’s it’s highly hetrogeneous but I I would love to study that yeah yeah okay so the bigger question so
Again thinking about the decision of the consumer um so I’m not only making an inference about the safety of a country I’m thinking about going to I’m like assessing the relative safety of all the countries and and if I’m making a decision like well I got to get I want
To get out of my cold country or cold place that I am and I want to go to someplace warm and perhaps you have a preference to go to the perceived safest place well you could imagine that if your sort of structural models correct those things are moving around quite a
Bit and so um uh I didn’t think this was a structural model but maybe it is if if I infer that a country is becoming less safe does that increase my probability traveling to a different country is there sort of like a reservation like well I just view the world as being too
Dangerous to leave and I’ll just tolerate you know the cold for a little bit yeah yeah I would again would love to study this because I think this is super relevant uh especially nowadays um think about the general equilibrium effects uh um you know both so the trade
Risk so to say you know there’s a effective exchange rate in relative risk um and there might be an absolute risk right or absolute risk perception um I think uh uh you know in order to study this we would need to have broader uh broader data uh which however has not
Been uh well it’s been just difficult to access because again some of this data as I described at the beginning uh this data was being sold commercially at very high high prices um making it basically un unaffordable which raises bigger questions around uh you know building knowledge public goods uh um and
Limitations there too um but I you spot on spot on I think both margin matter the extensive margin as well as the relative margin uh um but I think you know most consumers probably do in this case sequential decision making but that’s something that again people it
Would be great to test some of these uh uh uh mechanisms with micro dat you see some of the work from Leo burin and so on essentially doing exercises like this you with with experiments um which of course is great but it’s very nice to see this actually bearing out trying to
Map that to um uh you know data that um you know is let’s say you know stuff that moves gdps so to say so okay I want to ask another question on the more technical side you’ve got in both your reduced form exercises and in your modelbased estimates of the media multiplier you’ve
Got extensive sets of controls for fixed effects you got as I understand you’ve got diic fixed effects you’ve got origin by time fixed effects and destination by time fixed effects but I don’t see how you adjusted maybe I just misund missed it how you adjust for the fact that the Baseline
Level of news coverage is very different across your destination countries even for a given origin so there’s going to be more coverage of Israel and Egypt than Tunisia so when there is an event in Tunisia that involves coverage of violence or or crime against tourists that’s going to move things a
Lot um the proportion measure whereas if it was the same violent Evac the same implied increase in violence the Baseline Level news coverage about Israel is so much higher I’m guessing that it would only move your your true measure your your your truth measure excuse me your um your
Your newspaper uh your news coverage based measure a a small amount so that suggests there’s not going to be anything approaching homogeneous responses to the violent events yeah across the countries especially across the destinations is there some way your model is is dealing with that or are you just imposing
Homogeneity um so I guess on the reduced form I mean we we obviously check whether the results are driven by any one destination um and that that is not the case uh um the model essentially uh because you know we are you know it’s a the the the mark of switching model the
Filtering essentially ensures that you know basically what moves the needle in Israel is different than what moves the needle in Tunisia um and so I think that that addresses this heterogenity um that that you have in mind here uh um so uh um so on the reduced form side as I said
We can we can show that results are not s driven by any one uh uh destination uh um and through the modeling of the uh parameter space in a way that’s country specific uh um we uh we address again this concern uh um in a way that allows
Us to uh model this um you know with a un uniform pre-processing yeah okay but then the the media multipliers that you estimate the number three is that an average across all diad pairs the interesting heterogeneity is at the at the tourist origin level yes at the
Origin okay and so it’s which is exactly the example what I highlighted you know Russia versus Germany if there’s a violent event that is targeting a tourist in in in Egypt the media coverage that this gets in Russia is different than the media coverage that this gets in Germany yeah but I’m I’m
Still struggling there’s two things going on there’s the extent of coverage in Russia versus the UK about violence that happens in Egypt I’m I was actually asking about something different which is the Baseline level of news coverage about Egypt and Israel will be much greater than about
Tunisia how are you capturing how you certainly in the reduced form it’s analysis you showed us quickly it seemed like you just imposed homogeneity of the slope the response coefficiency yes yeah again on the on the reduced exercise you can just drop each country in turn you get similar coefficients so
That’s not that’s not relevant heterogenity I think in the uh in the context of um the model um I think uh um you know what we what we do is we have a destination specific arrival rate of news and so that takes into account uh the fact that uh places can be different
Sized in terms of some places just have a higher AR Ral rate or lower arrival rate I see um I think I think that tackles this in terms of the pre-processing so it’s a combination I mean my answer would be again to the uh the reduced form it doesn’t seem to
Matter that much uh um um and again because we allow for heterogenity in two ways through the arrival rate of news and through the um through the uh parameters the that that govern the you know the mark of change that these are destination specific um because we only
Have you know Morocco basically nothing is happening which is interesting in itself uh um um but obviously because we only had four destinations data only for four destination it was very difficult to kind of explore this further um so what we did was explore the heterogenity in the touris sending countries so to
Say and this is where the observation at press Freedom met us uh is I think quite quite insightful and quite revealing um which I mean in terms of the broader discussion of all of this uh work uh you know I very much tied this to for
Example the work on the that I did on the trade War um because there is a question to what extent uh um um countries that are not subject to well they do that do not have a free press are able to interact uh uh uh and engage let’s say in geopolitical domains very
Different because they’re subject to different domestic constraints which is also something that you know the media effectively imposes a constraint um which is also something that comes through in the paper that I did on the uh Coalition casualties in Afghanistan uh where is just a strong effect that where losing a soldier um
Has a multiplier effect and of course that opens the door for malign actors to you know offer bounties um um which is I think exactly what happened okay so um we got some questions in the Q&A first one’s uh very easy to to address so I’ll just take it
First um will this paper become available after the presentation well it’s already available I think on the we the the website for this Workshop there’s a link to the paper there but maybe you want to just tell us uh you’ve got a website where can people find the
Paper if they want to find it um yes you can find it on my website uh uh so TR Fetzer doc but you can also find it pretty much I mean just title if you look for the title of the paper the accepted version of the paper which is
Not the cut down version that we have to know uh um is available Open Access uh University of War has a Open Access repository for the accepted version of papers um so yeah you should be able to find it okay great and second question is it says does this multiplier idea
Work even if it comes uh in the form of good news uh so can positive news about a country generate um a multiplier effect if it’s on on tourism spending my hunch is it works for the denominator not the numerator um um just you know through a drowning out effect uh um so
You know um that’s that’s ultimately I think uh the main mechanism and you know what we can’t and what we did not discuss in this paper which I think is really important is the new selection function because there is reporting and then there’s reporting that gets put in
The evening news so to say if you think traditionally you have a traditional news diet um and I think um the selection function is uh actually uh really interesting there’s a nice paper I think that looks at the new selection function uh across sectors uh um I
Forgot I think it’s an it’s I think it’s an AR paper um that looks at heterogenity and the you know in sort of level business cycles and the new selection of sector relevant uh news that tries to quantify uh the relevance of the news selection function so I
Think uh um it does apply to positive news but it operates through essentially the drowning out of potentially bad news and of course there’s a dynamic component to it uh U the the buildup of a reputation uh which is we can also consider this to be a country’s brand or soft power
So one application I still want to do at some point is to bring bring this to bear on brexit uh in the UK because I have a bit of an issue with that okay with brexit okay um well you can elaborate on that if you want um but why don’t we
Give the Hans lters has a longer question um why don’t we let Hans um articulate his question we can we let Hans have uh have the mic for a second awesome yeah thank you can you hear me yes we can thank you awesome uh well
First of all thank you this was a really interesting talk um I really enjoy your presentation so I’m uh trying to understand uh kind of like how to interpret this like finding from the reduced form where you show that you know there’s a decline in spending after
This negative coverage and um I think there’s like two mechanism one is kind of like suggested by um by this like second finding that you have that um kind of the number of active credit cards goes which is just as the fewer people are visiting the place but I’m
Wondering if there’s like a complimentary second mechanism whereby um the behavior of visitors in the country changes as well um so for example like you know you have tourists that might just um decide to spend more time in their hotels rather than like going outside and spending money like
You know doing souvenir shopping and so forth um and I wonder if you have first of all I’m not sure how plausible at is so I would love to hear your thoughts on that and second um uh do do you have data on like the total number of
Transactions to maybe like get into the second mechanism as well um yeah great questions it’s funny I remember when I started working on this in 2014 I built this sort of booking.com scrape but I was going via the Wayback machine to extract essentially hotel price data uh
To look at some of these sort of uh some of these questions ultimately the data is to cause to allow us to look at substitution effects or behavioral changes within the country uh to distinguish um we cannot reject the null hypothesis that you know the difference in the average transaction you know is
Different from the of the effect sizes that we document in the uh you know on the Intensive and extensive margin so to me uh um like these are all really relevant questions uh um but uh with the data that we have we cannot study them of course I would think that this
Mechanism that we have in mind here applies very much to within country VAR ation as well think about uh uh uh um um you know behavioral changes uh around uh I’m not a good example comes to mind uh uh but you know any sort of new shock that might have heterogeneous impacts on
Media coverage that interacts with some people’s individual news filters or just their news diet you know where they observe or where they obtain their information from uh can produce heterogeneous responses um um and I think uh uh you know this is where you know I think this very much generalizes Beyond uh beyond
The specific case it’s just that for the C identification of course having an environment where we can segregate the information sphere a little bit better because there’s mechanic dividing lines simply language or distance and whatnot can give us this you know give us bite on this dietic uh dietic news measure um
Which I think makes this quite quite unique from an identification perspective great you know this thanks so much for this Teo I’ll give you a chance to make a few last words but it’s uh it’s very interesting work and it I’ve struck listening to you talk by how complex the
Information environment is um both for consumers trying to update their beliefs about you know where things might be dangerous or safe um there there’s no reason to expect that I can see to think that the um the actual dynamics of the evolution from safe to dangerous uh is the same across all
These countries so the the Markov chain May differ greatly across countries and of course that’s all being filtered through this um very imperfect reporting that is the source of information for many or most of the consumers uh so that the inference problem facing consumers is is extremely complex and challenging and then you’re
Trying to uh on top of that infer what they are inferring and how that’s affecting their behavior it’s it’s really quite a complex um problem when you think about it that way and and as we we’ve discussed there’s um there’s many other potential applications of your basic approach and again I the the
The health advisory one the Infectious Disease advisor seems to me like the most natural um applic of your framework and your methods I don’t know if you plan to to do that or or or done you want somebody else to do it yeah I think that would be great I
Mean to me uh uh I think some very relevant uh you know questions arise around just the geopolitics of stories um because ultimately when we are you know we’re moving into a more fragmented Global Order and uh a lot of things are happening that uh you know very difficult to comprehend But ultimately
Within this framework uh we can think of stories as being weapons of war in the service sector trade escalation uh um just like the decisions for countries to remove Travel warnings are something that that you know country leaders that Turkish president goes talk to uh talks to about
When he visits uh the German chanler for example so uh um these are materially relevant uh uh s from a geopol IAL side and there’s huge heterogeneity as I said like this this type of frame framework of thinking is just very interesting to measure for example soft power uh um other countries represented
Representation and in each other’s news and here is just a great example that I just put together for my year SE Grant application um this is three countries argentinia China and Germany what is the share of the top five news sources that that are represented so top five foreign
Countries and that that’s done with imperfect factiva data uh um but we see that uh for example in argentinia you know uh not surprisingly uh Brazil takes up a much bigger chunk of its uh reporting because of geographic proximity but then there’s Spain uh and there’s obviously a language Dimension
That matters if I think of the uh uh you know spread of uh uh let’s say um artists right musicians and so on where they have Market potential where they can tap in language is one of the factors that matters right um and so we see that there’s a lot of latent
Dimensions that I think are quite relevant to explain the Topography of news coverage or media coverage how other countries are represented each other’s news and of course uh as technology has evolved uh you know we’re move into a more Global Information sphere uh and the media consumption the
Type of how we consume news is often times you know generation or cohort specific Um this can create economic phenomena that economic as well as political phenomena that I think worthy of studying and we see you know like this a l more yeah okay great we should
Um sign off from the recording but we’re gonna continue the conversation for anyone who wants to stick around so thank you Teemo um this and thanks to the audience this was a lot of fun and uh it’s it’s been a pleasure to think about your work and we’ll see everybody uh next month
Bye-bye thank you so much Byebye