 I hope to not use all of the 15 minutes that I've been allotted. I think it would be good to have some time for a discussion a bit later. And it's been a real pleasure to read the papers for this session. So this is, I don't want to steal their thunder, but this is just a quick summary of what we're going to be covering in the presentations. So on US data about Facebook, we have a model that's looking to account for a very wide range of welfare impacts from economies of scale to the both profitability, but also consumer disutility of advertising to market power. And the question paper addresses is, what's the optimal policy in this case, allocating all of the ad revenues to users, building on the data as labor approach to digital advertising and data collection. We've got a paper using French data looking at the influence of Twitter on the production of news. How do news media of the conventional kind respond to popular posts online, both in terms of their editorial decisions, also economically in terms of their investments. We have a paper on Hong Kong data looking at posts and comments on online news sites. What does that allows to infer about conflicting viewpoints. And the short answer is that either people are in their filter bubbles and interact with each other or their fights. And I particularly liked in this paper the idea that there was such thing as a real bots compared to people who only comment once. Very nice approach using different styles of language to distinguish attitudes. Then we have a paper on French data again looking at a French merger and considering the consequences of a merger of a merger that went ahead, but the failures look at that as a two sided platform and the actual consumer detriment that came from keeping the ad sales houses separated. And then from David and myself paper on UK data looking at estimates of consumer surplus derived from these free in terms of money advertising funded goods, but also other kinds of free goods. The short answer being that interpreting the welfare consequences is a bit complicated as David is going to come on to talk about. On the face of it, although these are all about media and social media in some way, they are quite disparate papers and different questions different country data sets, but actually there are some common themes which is what I wanted to highlight here. One is that I really want to celebrate the empirics here, the use of new data sources, the creation of new data, and the addressing empirical questions that arise from these social media and media platforms. In my view, we've had a very high ratio of theory to empirics in this domain, and I think it's great that we're getting a lot of empirical work being done now. They also have a common motivating question, which is what are the welfare consequences of digital platforms, especially the social media platforms. And this I think is profoundly important and I want to come back to in my last slide. And then there's a common theme about the linkage between welfare outcomes and the ad funded business models. So I won't read this out, but in the papers, this is very common. Patrick mentioned my BBC experience, we did a lot of consumer research on behalf of licensing pairs. And one of the constant things that they said about why they appreciated the BBC so highly and it was highly is no adverts. So there's definitely consumer detriment from that and some interesting results in Mark's paper that he might talk about about the difference between adverts that compel your time because you're looking at them online or on screen and adverts on paper. In magazines and newspapers. There is an older literature about advertising. And I don't know how many of you know this Nikki Caldwell paper from 1950, where he walked through in an informal way the welfare effects of the advertising industry and came out with a pretty negative conclusion. He argued that advertising was a consequence of the market power around brands, as well as information provision, which is the general starting point in economics. You might conclude now that we have digital advertising, which is a huge market, highly non transparent, quite fraudulent with a lot of market power in the hands of Google and Facebook because the fantastic CMA study pointed out last year, that maybe we should revisit this kind of overview of, and perhaps not be as business model neutral, as we have tended to be certainly in competition policy. So, just a final few thoughts about this. And one is the policy implications. And this is a domain where I think the empirical approach is a really important because policymakers want to know, is this a big deal or not should we be doing something, how urgently should should we be doing it and what are the most urgent things to tackle. One of the obvious areas is competition policy. I was on the Furman panel in the UK. We know that there have been a lot of policy reviews of competition policy and now we're seeing that being put into action in different jurisdictions. So the lessons from the economic theory and empirics are being taken on board. I think one question I would have about that is whether it's sufficiently radical or not, but that's not today's subject. There's the question about news quality and the health of traditional news investment in journalism and the role of conspiracy theories and so on. And I think one of the things these papers open up is whether and to what extent in economic analysis we need to encompass these kinds of consequences. I should take it a bit further. Should we be worrying as economists about the consequences of anti vax conspiracies running rife on social media now. And there's also I think a question of working as economists with political scientists technologists ethicists to address these bigger questions about the role of government public service media. And a number of open questions that I think will come out through the presentations of the papers. So, for example, one issue that I've been thinking about is should we take a lens on welfare through time use. Because one of the things that the digital technologies and social media are doing is dramatically changing the way that we allocate our time both in work and in leisure. And a lot of questions about economic welfare generally, I think, taking externalities seriously trying to put some empirics on those, trying to think about how do we think about the very large individual consumer surplus that people derive from these goods with the societal costs that they seem to be imposing as well. We should end with a plea for us as economists to start taking welfare economics seriously again, including in the curriculum actually. I think we have to end the reservations that we've had as a profession about speaking to some very very large public policy questions and big normative issues facing our societies, which social media absolutely raise. I think you could say about the classic economic economists approach to economic welfare is that it's been a bit naive, politically naive, consequentialist, a bit naive about behavioral issues, and integrating these very different range, wide range of implications that the social media have for our society. So I'm going to stop that I really welcome these papers. I hope we can take welfare economics seriously in future work going forward. And I think the policy implications of the discussion that we'll be having will be large indeed. But if I do have any minutes left I want to save them for discussion later so thank you very much. So thanks for that great introduction and thank you for providing us this great opportunity at this most recent working paper with you. The title of the paper is called Anticipated Ideological Online Clash. It was a good political bias. This paper is caused with my friend Mandy, who she is working at the business school of the Chinese University of Hong Kong. So in Hong Kong, there are two major political parties. One is called pro-establishment party. Sometimes people are also calling them pro-Beijing or pro-communist party. They are the ruling party, and their opponent is called pro-democracy party. Ideologically, they are more towards western democratic political systems. Two parties are competing to each other. In March 2019, the HK government intended to pass an important law. A law is called anti-extradition law amendment bill that immediately hurts the feeling of the supporters for pro-democracy party. So they go on the street and they do a lot of protest. As you may see in the picture, during the protest, there are a lot of clashes between police and the civilian authors that makes the word police becomes the focal point of the ideological conflict between those two parties. Basically pro-democracy citizen, they think the government has abused the power of the police whereas the pro-establishment supporters, they support the government's use of police force to maintain the social public orders. The discussion about police are raging from offline to online and that motivates us to write on this paper. We simply collect the data from Facebook's HK mainstream media website. Our data curves one year from April 1, 2019 to March 31, 2020, that includes more than 140,000 media reports from 44 media outlets. So we focus on studying those comments under each news report. There are in total around 40 million comments. The one interesting thing is that almost all the comments in our data are provided in the use of Chinese characters, but there are two ways to write on Chinese characters. As you may see here in the picture, this is the word in Chinese character, meaning knowledge. On the left-hand side, you see expression under the usage of simplified Chinese characters. On the right-hand side, you see expression under the usage of traditional Chinese characters. The difference writing habits was due to the fact that the Chinese Communist Party, they laid a big reform called Chinese character simplification reforms during the 1915s. So as a result, those people living in the mainland of China, they used to use simplified Chinese characters whereas people living in Hong Kong, they were being educated since their childhood to use traditional Chinese characters. Both people can understand each other, but this is bit costly for those people to switch their writing habits. So different writing habits in our paper also represent the basic culture and the original differences amongst those people. So in this paper, we mainly want to study what's the impact of the number of populist comments on the number of anti-police comments. Specifically, under our circumstance, the number of populist comments may generate some further externalities. Imagine that on the platforms that the voice has been dominated by those populist comments. It may urge the government to use more police force to maintain public order, and that may also enhance the risk of offline protests. So such cheating effect may deter those poor democracies supporters' motivation of expression they are feeling online. So we want to check if there is such kind of a cheating effect and I'm going to show you the result. On the meantime, you also want to explain what are the factors that may intensify the online ideological conflict. If we're pulling on all the comments together, we see that those words that have been most frequently used are like police support and black cops. Those words supporting police usually and often appear in the same comments, and the black cops is a term that has a strong feeling of anti-police. That confirms two things. The first thing is that as we mentioned, the police is the focal point of this ideological conflict during the last year. The second is that those people who are supporting police and those people who anti-police are evenly matched online. So in the economic parts, we use the dynamic panel regression model to study the impact of proper police comments on anti-police comments to check whether there is a cheating effect on people's willingness to express their demand for the Hong Kong independence and autonomy. We also include the words that is the number of the comments included in the words of Liberal Hong Kong because Liberal Hong Kong is considered as the slogan by those protesters and the poor democracy supporters during the last year. That has been widely used. To deal with the indogeneity issue, we propose a new set of instrumental variables, and later in the empirical part, we find those instrumental variables, at least in our circumstance, perform pretty well. They pass all the tests related to the validity of the instrumental variables. The idea is to include the number of new active simplified Chinese users as well as the number of new dead simplified Chinese users. The idea is that those simplified Chinese users used to support police that satisfy the relevant conditions. In the meantime, no users as they ask can check these users' historical activity and their online trajectory. So that makes their exclusion restriction condition may also be validated in our circumstances. Let me just briefly summarize our estimation results. The IV estimation shows that for each one proposed comments that are going to lead to more than one anti-police comments. That is very important because such kind of impact dispels the cheating effect and leads to more online demand for autonomy. We also manually collect all the offline protests that happened in Hong Kong during the last year. We do the regression model, so we also show that there is a strong link between people's online expression for their demand for the HK's independence and autonomy and the frequency of the offline protest. Beside of that, we also have two interesting findings. The first thing is that when switching to those comments specifically provided by those simplified Chinese users, it seems that it intensified ideological conflict. The second is that all the estimation results consistently show that our estimation may underestimate the actual impact. So we do some further check. The first thing is that we check why this may intensify the ideological conflict. We find that so the ideological conflict might be intensified just a minute due to the use of different writing habits riser from those people who provide the comments that have different cultural background. So that indicates that the reform of Chinese character simplification in the 90s lead to a very long standing cultural bias and that's intensified the ideological conflict that we observe today. And the second thing is that we surprisingly find that many comments might be provided by those people who are considered as suspected boss and the existence of those comments may weaken the intensity of the conflicts. So the last thing is related to the policy suggestions. We have two things to say. The first is related to the China's great firewall policy that has been long time criticized by many people saying that this policy basically that is people living in the mainland of China freely browse the foreign website. And actually we find that actually the effect of the policy is two-sided because on the other hand, it's also can drastically weaken the ideological conflict between China and the US countries like Western Wars like the US. The last thing is related to the water armies because many in government have the intensity to use online bots and the water armies to do the ideological propaganda. So we find that in terms of awakening the online conflicts, such kind of policy might be effective since we find that there's some evidence that shows that low quality water army may play down the conflict by diluting those conflict generating content and high quality water army that interact more with other people. We play down the conflict by preaching people. So this is all about our results. Thank you for your attention. So while you're presenting already, so here we am. Thank you very much for the opportunity to present this joint work with Diane. Here we're going to talk about free goods and economic welfare. I'm going to try to give you a brief overview of the of the paper. And I hope by the end of my eight minutes, you all go to the ESCO website and have a look at the actual working paper. So let's get into it. So basically, our starting point is that I think we can all relate to that, not only this year, but in general that our personal and professional lives are increasingly moving online. So here's some facts about UK, for example, around 90% of the population do use the internet regularly, more than 80% have a smartphone. Many people actually now is only access the internet via mobile devices as opposed to desktop computers. Now this is important. So UK adults spend around three and a half hours on average on the internet every day and 70% of those of this time on mobile devices. Now, by the time of the height of the lockdown, this has to increase to more than four hours per day per adult on average in UK. And as you know that many services that we use online, some of them I listed here, Facebook, WhatsApp, Instagram or various Google services, not Netflix, of course are, but the others are mainly free to use in monetary terms. And that's going to be important when we think about what's the consumer surplus and how well do traditional measures of consumer surplus do. Okay, the context for the study is simple. How do we measure the value of an increasing number of goods that are free at the point of use, even though I started earlier. We have a couple of goods in here where in pay, such as Spotify or Netflix and other services such as public parts, we don't pay directly. And those are offline and we trying to work out some of the comparisons here between on and offline goods as well. Of course, this is not new questions valuations in the absence of market prices has been a question that's around in the environment or culture, and obviously that's there's a large literature on all of those. Now, Eric Rinalson, Abi Collies, and Felix Eggers have used these contingent valuations techniques then to find really large consumer surplus from free digital products in a paper published in 2019, and some subsequent studies and papers. And that's really our starting point for for a lot that we'll be doing here more broadly. We really tried to think about this kind of spectrum of GDP and welfare and you know where, where are we with with certain measures and how good. How good are they do we need to think about creating new ones if GDP is increasingly challenged by the digital economy. We're able to use the UK lockdown to assess the changes in these valuations because quite suddenly you could not use a lot of offline services, and we're forced to use to use online. Okay, we conduct conducted so far two ways of an online survey why you got around 10,000 individuals were asked in late February 2020. So that was for the lockdown in the UK and elsewhere, and quite at the beginning of the whole pandemic so in that sense, we're quite confident that our results are not really driven by that at least and we can see quite significant changes for example if you repeated that 10 weeks later, that you know a lot of the usage has changed across different goods and digital goods, and also the associated valuations as I'll get to in the next slide. We're planning now to run the survey again in Iran in February 2021. So what we're doing here is we're asking around 30 winners to accept questions of this. Here's an example where we're asking people how much or what would be the minimal amount of money that we're willing to accept to give up a certain digital good and that there's a whole long list of those. And this is the example for for Netflix. For example, we know, of course, that the consumer surplus will be higher than what people pay for it randomized to question order randomized time periods so we randomly ask people to give up the good for one month 12 months. So we can play around with some of those later on. Now, of course we have 10,000 individuals so we can kind of calculate the average willingness to accept for for our sample because our sample is weighted by various characteristics to representative of the UK adult population. Of course, it's important to point out that the adult population is only the adult online population. So, as I mentioned in the beginning, around 10 11% of the population is not online so we need to think about how to account for those for now, I would say this is representative of the online population. Now I've just picked out three examples of the 30 that we have. The average valuation for Facebook is around 1200, almost 1300 pounds per year. It's higher for public parts, it's almost 2000 pounds. And for Google search or online search, it's more than 3000 pounds a year. These are average figures, we know that some individuals have very high values. They look at millions, they obviously much lower. All of those are in high use, so around three quarters of the online populations and almost 100% in the terms of search will be using these. Now we can, we can use these obviously to plot some sort of demand curves, and we can say, okay, if offered a certain amount of money, how many people are willing to, what's the share of the population that will be willing to give up access to those for all the willing to accept thresholds that we have, then we get to these demand curves, we can do the same in May, and we're going to do the same now in February 21 to kind of see how, how the demand curves shift or tilt, and whether the demand for some goods is becoming more or less elastic or more or less elastic over time. And I think that's quite quite some interesting insights already here that I don't have time to wait into. Okay, we can look at non usage rates and valuations and you can see this, this is kind of non linear relationship between, you know, how much, how widely spread or in use certain good is and how high evaluation is. In general, there's a that case because it's not usage. It's a negative correlation. But the important thing is non linear. So we think there might be some sort of network effects at play here. Okay, I'm going to skip forward a little bit in interest of time. We have individual characteristics so obviously we control and kind of work out different really interesting differences across age groups or gender and income and regions as well. All of those are in the paper. We look at the impact of COVID we kind of how that has changed. And then we found two issues here one is time inconsistencies. So one issue is that if you add up, or if you multiply the monthly valuations by 12 you don't really get the annual valuations which is an interesting thing to bear in mind. And we struggled with the budget constraints. So if you're not paying for something, what is the real budget constraint. And that has to be something around the value of time so that's something that we're doing some some further work on at the moment. So conclusion is that we think that these online service are highly scalable so actually they are viable tool to change it to elicit these changes in valuations. It's easy to expand the categories add and drop products, but generally the willingness to accept is very high. And they also differ across social economic groups. I'm going to stop here. Thank you. Our next speaker is a person who will present a joint work with him in Australia on no less than how to govern Facebook. All right, excellent. Can you guys see my screen. Yes, excellent. Okay, so thank you to David for that excellent lead in because what this paper does is really take those kind of willingness to accept offer demand curves that you guys estimate in your paper and obvious my colloquial. My co-author has estimated in the past and try to take those demand curves seriously and plug them into a structural model of Facebook. So let's see what happens when you do that. So first off, we all know that we're kind of worried about monopoly power amongst these big firms, we think they have supply side economies of scale because of low marginal costs. And for digital platforms, we also think that they have demand side economies of scale because of these network effects that make the product more valuable as more people use it. People have talked about different ways of dealing with it. Europe's talking about taxes of different kinds. This data as labor idea suggests that maybe we should unionize users of these platforms and have them collectively bargain against Facebook for their share of the profits. You can also imagine doing things to improve competition. Fiona Scott Morton in her call for interoperability is a big part of that. Maybe we need more competition in this space. And then some people say that the way you get competition is not so much by interoperability and regulations, but just to splitting up these companies and you can imagine a vertical split up where for example Facebook gets split up with an Instagram which is a very if you think of Instagram as being a different market, or you might imagine like a horizontal split up where we end up with two baby Facebook's kind of like how bell telephone was broken up. Okay, so just because it's a short presentation the result preview in case I don't get to everything. We find that Facebook generate seven times as much social surplus as it collects in ad revenues in the United States. That's the good news for Mark Zuckerberg. And Facebook's market power lowers welfare by 4% versus perfect competition, but what matters would counts and social welfare is actually going to matter for that right. What we're going to find is is that Facebook actually has a lower level of advertising versus what it's kind of short term profit maximizing strategy would be. And from that we impute that they have a value for maintaining a large user base. And if you think that that value is from their collecting data that's going to help them make new products in the future. Or plus that we should count. But if you think that valuing a large user base is more about a mode and deterring entry, then that value of a large user base is maybe something we shouldn't think about in social welfare. But I'm going to in these presentations I'm going to take the positive view. Taxes we find are going to be mostly interested on Facebook, but a properly targeted tax can actually raise consumer surplus and so that's going to be making raising taxes a pretty attractive solution to a redistribution problems especially if you don't have a lot of domestic investors in your country. We find that a network effect killing breakup would be disastrous hurting producers and consumers, and then data is labor this idea of rebating to customers, some share of the value of the ads that they watch is win win because we preserve the ads, but we're also incentivizing people to use the platform and get the positive network effect. Okay, so what does the model look like again don't have time for details I encourage you to look at the video and watch the paper, revise versa. The two kind of types of agents in our model or the platform firm, which are trying to try and to maximize its objective function, which is profit plus maybe also they value maintaining a large user base. And then you've got the consumers, the consumers have to make a decision about whether or not to use the product. The consumers are heterogeneous, which is why we call this a multi sided platform, you know building on Jean Tyrol and all that great stuff, because each different demographic group of users or whichever group that Facebook is price discriminating at against is going to have create different network effects for every other side. The grandmas are going to really love their grandkids being on the platform, and maybe grandkids don't care as much about their grandmas being on the platform. As well as differences in opportunity costs and differences in opportunity costs are going to drive differences in those demand curves, like we saw with in David and Diane's presentation. So that's the model in order to solve the model computationally we solve it through a series of cascades. So we asked the question, suppose that Facebook were to raise the level of advertising on group number I group I, well then people in group I are going to use the platform, somewhat less. And then so on and so on you have in sort of an infinity of these cascades until the model reaches a new equilibrium and nobody wants to leave anymore. Practically we find that the model kind of dampens out after about three or so cascades. Okay, so now calibrating the model for Facebook. Again, similar to Diane and David, we're going to go out there and do these online surveys. We're going to estimate these online surveys for 12 different demographic groups and try to get a sense both of the sort of overall demand curve for Facebook, the value that's coming from users of different types. So what what kind of users do you care about being on Facebook. And then finally people's disutility from advertising and we also bring in other information from Facebook's ad API and their quarterly reports on basically revenues and disutilities. Okay, so this is our demand curve. This is very analogous to what you saw in the last presentation. These three dots are the dots that are in our survey. So this is a 95% confidence interval. And again, it's a demand curve is how many people stop using Facebook if you give them $20 $10 $50. And again, similarly to what you saw before, we find the median is going to be less than mean. So it's inframarginal value is going to be very important in this setting. This is the network network effects. The size of the node here is the number of the initial users and the thickness of the line is the value. These are the most valuable connections. Like I say, grandma's really like grandkids. Grandpa's like middle aged men and middle aged women like being connected to older men. Grandma's got a lot of value from everyone provide less out. Kids get less value from everyone but provide a good value. So this is heterogeneity is going to be important for understanding how Facebook's going to respond to different policies. So like I promised you, we were going to simulate six different policies. Here are anti trust simulations. These latter two represent two different kinds of breakups horizontal breakups. So we're left with two mini Facebooks. And then a vertical break up where we take Instagram off of Facebook and we just sort of arbitrarily assume that that makes main Facebook quality. About 5% or soft. As you can see, both of those policies are going to reduce the amount of people on the platform and reduce quality and make people unhappy. Whereas perfect competition. If you're imagining what this represents is advertising coming to zero but no killing of network effects perhaps through Fiona Scott Morton's interoperability. We see that that would obviously kill Facebook's profits, but would increase consumer surplus significantly and lead to a boost in social welfare of about 5 percentage points. Which is like a lot given that we're talking about a lot of value going around here. But maybe tax and redistribution is the better set of policy tools to use. So we evaluate a 3% tax and now remember because Facebook has these two motivations, the large user based motivation and the profit motivation. If you tax their profit motive, they're actually going to substitute into their other motive, right? So that's the mechanism by which a tax on advertising revenues or corporate profits can actually enhance consumer welfare. Whereas a per capita tax, a tax on the number of users that raise the same amount of revenue would have the opposite effect. It would slightly lower the amount of users in social welfare as Facebook squeezes the most elastic users tighter. Whereas data is labor is kind of a win-win, you can just pay users for viewing the ads. You can imagine a whole range of redistributions. This is an extreme case where everything was rebated and that has a big positive effect on social welfare 30%. Finally, why just to tie this into current debates, why not tax Facebook if you have few or no domestic investors or innovators, right? As France, if none of you own a Facebook stock and I'm telling you that you can tax Facebook and have no effect on consumer welfare or even a positive effect on consumer welfare, why not do that? Well, Proposition 22 in California, very briefly, there was a law that was passed saying that Uber and Lyft had to basically give more benefits to their drivers. Uber and Lyft said we will leave if you make us do that. There was a proposition and California backed down. So perhaps the ultimate limit on these taxes isn't social welfare, but just bringsmanship against these firms. So thank you so much for your time. Call to action, we need more people collecting this data and building structural models. On this encouragement, thank you. Our last speaker is Mark Ivaldi, who will present some long work on with the design on trying to draw the lessons from a case on the TV market for understanding platform measures and sometimes unintended consequences of some of the remedies that are sometimes used. Mark, the floor is all yours. Do you see my screen? Perfect. Okay, thank you. So this is a joint work with GK Zang. I think that our paper is complement to the last two one in the sense that what we are going to try in some sense is also to have an evaluation of the value of TV show. When the only price is in fact a negative price through advertising. So this work started with a decision from the French Authority. The French Authority has approved the acquisition of two small TV channel by the biggest TV channel in France. But in the same time it had imposed a remedy that is to say to keep the advertising sales house separated advertising sales house on the part of the TV channel TV station that is in charge to sell slots of time to advertiser for their advertising campaign. So the basically the rational behind this decision is relatively straightforward on the book at the inside there can be there cost efficiency in the sense that the free channel are going to share the same catalog of videos of TV shows. So that explains why the authority has approved the merger on this side. On the other side, if the authority had approved the merger of the advertising sales house, it will have reinforced the dominant position of the biggest TV channel on the advertising market in France. And so the remedy. What we can say from that is that clearly the two sides have been treated separately. At least we don't see in the decision a relation between between the two to decision and is for this reason we we thought it was interesting to evaluate this decision because it is indeed a very interesting market because on one side viewers watch TV shows for free. And in some sense pay by receiving flows of ads. By the way, it is very similar to looking for information on the internet where you go to Google to find an information is for free but you receive in the same time advertising. And on the other side, indeed there are advertising generate revenues to to to the to the platform. And as you know, the difficulty here, the challenge is to understand the relation between the two types of externalities that you can have here. We have positive externality of user on advertisers. Since it's free, free of charge to watch TV. Of course, the more user, the higher the incentive to increase the quantity of advertising on the other side. It's not always true, but in general advertising is perceived as a negative externality. And so there is some incentive to decrease the quantity of advertising. What is what is equilibrium solution of this bit that results from these two forces. It is the basis of the structural analysis that we have tried to set up. So it is not easy to get data on advertising. I mean, it's a real issue to analyze all those markets. If we really believe in this type of model of two sided, we need to have a data on both sides. And that's the big challenge. So we have been able to get monthly data from the period 2008, 2013. It's monthly data. Me, we, we, we, we could prefer to to have more disaggregated data to to analyze more precisely the decision on broadcasting, for example. But anyway, this is interesting because we have been able to observe what happened after the merger that happened in 20 in 2010. The data covers the major free boss cast TV that cover 90% of the total advertising revenue in France. And the free type of information that we have, and that I needed to, to, to analyze a model of supply and demand is that we have an information on on the number of the viewers in term of a weighted average of viewers per second in in a month. And we have any for me any formation about the price of advertising. We have also the quantity of advertising, and then which is important for instrumental variable information is to, we have the data on the broadcasting content of featured TV channel in each in each month. It's on the on the on the market structure. So I already said that we are going to consider TV station as a two sided platform because they provide two services TV show to viewers and advertising slot to advertiser. The real question is of course to evaluate the sign and the magnitude of the network externality between these two type of customers. One important difficulty here is that advertiser practices a multi oming strategies. That is to say, there are different, we can think about that in a different way, but the idea is the objective of advertising is to minimize the cost of advertising campaign by choosing the mix of TV channel where they are going to put their their advertising slot. So, so, and we have to address this, this important, this important question that has not been very much studied in fact on an empirical part in the in the literature. An important, an important aspect also is the type of competition that are between TV channel. We, we believe in fact we tested it also is that the Cournot behavior is the, the, the most realistic concept to to represent the competition between them because in fact TV programming is defined maybe months before. And then the advertising sales house determines the supply of advertising slot that they need to take into account some regulation imposed by the French law and of, and they take into account the sensitivity of viewers to advertising. So, so the release is it's not to compete to our part of this, not a competition in price, but a relay in competition in, in term of of quantity and the Cournot behavior works very well. We did it, we did test on that in the paper. Okay, so sorry. Can you walk up in one minute. Okay, so the model is very simple. There is a demand model estimated by nested logit model for the for the viewers. We estimate the demand of advertiser using a translog cost function. And we are able by this way to to analyze the substitution between TV channel from the point of view of viewers and the point of view of advertiser. And then we are able to use our estimation of demand to to analyze the equilibrium between the TV channel. So here I put the equation as you can see the price cost margin of TV channel is very complex. It does not only depend on the demand elasticity in term of price, but because here there is no, there is no price to to to access TV show. So, but we have to take into account the effect of advertising on the price of advertising, the effect on the viewer. Number of viewers on the viewership globally and our advertising affect viewership. So the solution of the equilibrium is quite complex. So, given that I don't have too much time, what is interesting is that we have simulated we have observed the equilibrium first post merger under the remedy of the of the French authority, which was, as you remember to to keep separated advertising sales house. And we have simulated the merger without remedy. Okay, in both cases, the consumer surplus decrease. So it means that the decision has not been right. If the objective of the authority is to preserve the consumer surplus on the side of viewers, then the decision has failed. Moreover, if they did not have impose the remedy, the profit of TF1 will have increased. What is interesting here is that in fact the decision has affected the other party of the market and in particular has increased the profit of the other TV channel. So, so, so to conclude, rapid lay on the is to free conclusion on the basis of consumer surplus. The merger should not have been approved on the basis of the welfare analysis. The effectiveness of the remedy can be discussed and the main lesson is that it's very important to take into account the two side of platform. Otherwise, we obtain results that are not not really expected. Thank you very much.