 Mae'r prins yw'r bau chyfnodd, ac mae'r ffordd o bai allan pethau. Mae'r bai llwythion ar gyfer y cael y sgiliau, ac mae'n gwybod eich bod yw cael eu cwestiynau i'r llwythenau, mae ydych yn adil i'r perthynau yn eu cyfrancau ddau. A bobl hwnna wedi a ddweud cael ei ddechrau o bai allan, mae'r cwestiynau i ddweud cysylltiad i chi'n wneud y mhaneu ar gyfer y maedlau i'r that could lead to concerns of foreclosure. So this issue has been cast into the limelight in the world of technology, where of course we have many intermediaries that are making recommendations, providing search results, thereby have a lot of scope to steer consumers in various ways. And so we've seen a wide range of cases in Europe, the US also globally, and also particularly in the last year or two, there's been gathering momentum behind a wider debate about how we should be regulating these gatekeepers. And so people have started to talk seriously about interventions like structurally breaking up the intermediaries, or at the very least imposing some sort of neutrality obligation, for example, on them. And so this is really a germane topic, not only for economists, but also for practitioners. Of course, whenever we have something that's a policy interest, there's inevitably an economics literature that follows it. I think by now we have a fairly mature literature that helps us to understand when would we expect these sorts of bias to arise, and what sort of effects might we expect them to have on market outcomes. And this is just a selection of papers from now, a relatively large literature. But this session brings together papers that are a little bit closer to the research frontier in a couple of respects. So firstly we're going to see some empirical papers. And I think it's fair to say that we have a slightly less developed empirical literature in this area than we do a theoretical one. And so that's going to be quite exciting. And then also we have four papers, theory papers, and they are going to look at a particular aspect, namely the choice of platforms about whether they enter their in marketplaces and therefore what sort of business model they're going to operate. And that's a particular issue that's been subjected close scrutiny in the last few years. So to start with, we have two empirical papers. I want to stress that I think this empirical work is particularly important because as is often the case, theory has been very good at highlighting some of the trade-offs that these platforms face. Often those trade-offs boil down to an empirical question of, you know, is the platform going to prioritise its short-run interests or the long-run growth of the ecosystem, for example. And so I think it's quite important that we start to get some rigorous empirical results that answer those sorts of questions. And so first we're going to hear from Ulrich, who's going to present some evidence about self-preferencing in the market for online travel search platforms. For each of these papers I'd like to draw out something I think is a little bit interesting that you might want to pay attention to in particular during the talk. One thing I found quite neat about Ulrich's paper is that there's really a lot of nuance to exactly how these websites are designed and what implications that has for self-preferencing. In so in particular, we have to be a little bit smart about identifying exactly what sort of behaviours correspond to self-preferencing, but that also presents an empirical opportunity to find new ways of measuring that kind of behaviour. So there's some pretty interesting evidence there. Tiffany will present a paper in a similar vein, this time looking at recommendations on Amazon's website, again looking for evidence of self-preferencing. It's a neatly controlled study, I think, and one of the interesting things about this environment is that as well as looking at the advice that Amazon is giving consumers, we can also get some measures of the effect that that advice has on consumer demand, and that allows us to start to also think about questions like whether Amazon is really responding to short-run incentives in the way that we might expect it to when it comes to self-preferencing. So I put together a few, what I call here, big picture questions, big picture in the sense that they're not aimed at any of the papers in particular, but I think these are some of the issues that are important in general to think about when we reflect on this broader literature. And so the first of those is what is the motive of the platform? And so I already hinted that, you know, in the short run, they might have a lot to gain by participating in their own market, steering all of the consumers to their own products. But that might not be the behaviour that maximises the long run value of the ecosystem as a whole. And so I think there are really questions about which of these incentives is the platform responding to, are we really seeing a kind of anti-competitive behaviour, or is this in fact a sincere attempt to try to provide the sort of service that is best for consumers as platforms often argue when these things come up in cases? And so I think getting a good empirical handle on those sorts of questions is really quite important. Secondly, of course, we can never study everything in a single paper, and I think it's always important, of course, to ask what are the things we're not studying, and I think issues like the quality of products that are available on these platforms and investment in them, the decisions of third-party sellers to enter the platform or not, and also the wider decisions that the platform is making about how interactions on the platform are structured and how that's governed. I think these are all important issues. There's a bit of literature that looks at some of these, but I think there's still a lot of scope to do exciting stuff empirically in this area. And then lastly, once we identify evidence for self-preferencing behaviour, there are a lot of interesting economic questions that flow about the strategic implications of that. How would that distort the behaviour of the third parties on the platform? Is it going to change the kind of investment decisions they make, the way that they themselves set their prices? What implications does it have for the growth of the platform and the competition between platforms? And again, I think there's a whole host of interesting questions to be answered there too. After those two empirical papers, we move on to four theoretical talks. These focus more specifically on the issue of entry into a marketplace by the platform that is running that marketplace. That's really been one of the linchpins of a number of cases that have been going through the system in the last couple of years. And it's a fairly challenging problem, I think, to think about because, as we all know, these platforms are complicated environments with a lot going on. You have different types of agents interacting the fees of the platform, the prices of the agents trading on the platform and so on. And here in particular, we have four papers that are going to shed a light on four interesting different aspects of this problem. So we're going to start with Ursulam, who has a model of entry into a marketplace by the platform that runs that marketplace. And this is a paper about the strategy that the platform uses for entry. So how is it going to set its price? How is it going to choose the fee that it charges third party sellers? It may come as no surprise that those decisions are going to have an impact for competition on the platform. And something that's interesting about this paper is really that we get a chance to see how those effects interact with the quality advantage that the platform might have. And that's going to turn out to be really economically quite important. And also look out for a fairly neat methodological contribution in this paper. I think it's a neat way of approaching a difficult problem. Then we go on to Leonardo's paper. It analyzes a very similar kind of problem. So we have a platform that's entering its own marketplace. But Leonardo's going to focus on the important subset of cases where the products that the platform sells or that are sold on the platform are subject to network effects. So I think of something like smartphone applications, for example, where we think network effects are likely to be quite important. And so what Leonardo is going to show us is these network effects have quite important implications for the consequences of entry by the platform. And then we have two theory papers that look more at market dynamic issues starting with Winspaper, which looks at the issue of data use by the platform. So this is an environment where the platform can take data from third party sellers and use it to figure out what kinds of markets would be good ones to enter, what kinds of products should I consider selling on my own platform. And that's important from a policy perspective, but it's also going to turn out to have a range of interesting strategic implications as sellers try to manipulate the types of information that are available to the platform. And then we'll finish up with a paper from Joe looking again at dynamics. And so Joe is going to look particularly at the relevant case where the platform is not only making money through transactions on the platform, but it's also perhaps selling a device that you need to use the platform. And so it's thinking about the revenues that come from that part of the ecosystem as well. And so what Joe is going to show us is that the long term dynamics of those device sales is going to be important for the strategic decision of the platform, whether to enter or not. And there are going to be various kinds of commitment problems that face the platform in these sorts of environments. So again, just a few big picture questions with more of a focus on the theory. I think, as is common in this literature, there's because there are so many different issues at play here. There's a bit of a challenge in thinking how can we package all of these insights together into a sort of coherent recommendation for policymakers about how they should make policy in this sphere. Secondly, related to the point I made in the empirical part of the introduction. I think it's important to think about how we weigh the long and the short run motives of the platform. We may be able to identify in the short run that there's a clear incentive to bias. How can we also account in the models for the long run interest the platform has in its own ecosystem. And really, you know, often to make our models tractable, we're going to simplify by focusing on some aspects of the problem. And that naturally begs the question of who are the actors that we're not thinking about and what are the strategic decisions that we're not thinking about as much. And so again, things like investment by the third parties or other sales channels that they may be considering competition between platforms. Things for us to have in mind as we try to come to holistic policy recommendations in this environment. So I think I will stop there and leave as much time as possible for the talks, which I'm sure can be very interesting to listen to. Thank you very much, Greg. The first speaker is Uli. Uli, do you want to share your screen? Yep, go ahead. Okay, so thank you very much for the nice introduction and having me. So it's my pleasure to present to you ongoing research entitled Vertical Integration of Platforms and Product Permanence, which is co-authored with Morgan, Matthias, Reinhard and Thomas. So online platforms are an essential part of e-commerce and a core function of them is to provide recommendations to products but also to suppliers offering these products. At the same time, it's common that online businesses have ownership links between them and these ownership links are a major concern in competition policy, as Greg also has discussed already before and also has reflected in the many European cases we have right now. So we would like to contribute to this debate by studying the link between meta-search platforms and vertical integration in the hotel booking industry. The hotel booking industry is an interesting case because it features many online platforms with complementary business models like meta-search websites such as Kayak allowing the comparison of prices for one hotel across different sales channels. But then there are also booking websites for reserving accommodations such as booking.com. And as you can see here on my slide, Kayak and booking belong to the same company group as many other online travel agents. And this is the booking holding. Okay, so to explain a little bit what we are concerned about. So in 2013 price line acquired the meta-search website Kayak for 1.8 billion US dollars. And price line, which is nowadays the booking holding, consisted already back then of many online travel agents like booking.com. So this raised the concern that Kayak would not be neutral anymore in its ranking algorithms or in his recommendations after the integration, but have the incentive to promote affiliated online travel agents only. To which Kayak announced it would not buy its search results. We can exemplify the concerns that were raised already there by looking at the website of Kayak. So Kayak allows finding hotels on different sales channels. And you see here the search results for an overnight stay in Berlin. And Kayak presents us with a list of hotels and some sales channels for these hotels. So in our example, we have two hotels. And for the first hotel Kayak shows most prominently the offer of Agoda for 58 US, which is a booking holding platform. Kayak also shows less prominently the offers of Expedia here and hotels.com and ebookers. And then there are some three other offers which are hidden in some dropdown menu. And given here that Agoda is offering is one of the platforms or OTA is offering the lowest price. One could say this horizontal ranking of sales channels isn't the best interest of consumers. If we look at the presentation of the second total though, things are a little bit different. So Kayak chose here the OTA booking.com most prominently. Although Expedia, which has the lowest price here or is one of the platforms with the lowest price is showing less prominently here. And however, this should only be the only thing that you should be concerned about. So in our stylex example, it's also the case that Kayak shows a hotel on the first position, which has the best price on a booking holding platform. While it shows a hotel with a lower price on a competing platform like this one, the Radisson Blue, which has a lower offer on a competing platform is shown later in the search results. So by this, you see that Kayak potentially has two means to steer consumers towards their own affiliated platforms. So this is first the position of the sales channel for a specific hotel in what we call the horizontal ranking. And then second, also the position of the hotel in the vertical ranking. So here the order of the hotels. And this is what we essentially study in our paper. So in our paper, we address the question whether the Metasearch platform Kayak favours their own affiliates like booking.com. And for this, we use web script data from Kayak over a period of three years. The data covers 1,600 hotels in Paris, which lists prices on up to 22 sales channels. And by this, the data set consists of roughly 16 million observations of prices, but also of ranking decisions of Kayak. So what we first analyse descriptively is how often Kayak promotes sales channels of distinct groups and how this compares to the price structure. And looking at cases where we have a unique price leader, which account for 72% of all offers, we know that Kayak puts affiliated OTAs more often prominently than they are price leaders. So if we, while booking.com, while booking holding OTAs are price leaders in 10% of cases, we see that they are put most prominently in 22% of cases. Of course, this descriptive analysis does not take into account that other sales channels might be less popular. So Kayak might take into account, okay, actually people like booking for instance, and therefore they promote booking more often. And therefore we investigate this further by conducting a regression analysis about the ranking decision, taking different measures for popularity of sales channels and hotels into account. So the regression analysis we do first concerns, first concerns the horizontal ranking, so which is the decision which sales channel Kayak is promoting or making visible. And there we note that Kayak makes offers of competing OTAs significantly less visible, and is showing them significantly less often prominent. So anything else equal offers of Expedia OTAs are 11% points less visible, and 7% points less often prominent. Second, coming back to my initial slide with a screenshot, we also study the decision of Kayak about the vertical ranking. So this is how Kayak is sorting the hotels in the search results. And while interesting here the price itself does not seem to be very important for the decision how to rank a hotel at all. We observe that it matters which online travel agent is the cheapest one. So specifically compared to offers of to other offers we see that hotel offers where an OTA of the Expedia group has the lowest price. They receive very significantly worse ranking position in the search results. It becomes even larger once we control for variation in time of the popularity of these hotels and when we control more for the popularity of sales channels because as you see we have an observation period over three years, such that we can benefit from a lot of variation in the popularity also. A part of the caveat of our analysis is that we do not observe the commissions page to Kayak by the sales channels, and Kayak of course takes them also into account when deciding about the ranking. This can of course be a problem if there's substantial unobserved heterogeneity between hotels over and over time about these payments. But we are confident to some extent that some sort of hotels and some sort of sales channels on average to this advertising similarly. But still to mitigate this concern we conduct a comparative analysis on a different meta search platform, which is not integrated with an online travel agent. And for this we choose Google hotels. And here with respect to the vertical ranking, we don't see this pattern anymore as on Kayak that hotels which have lower prices on Expedia receive worse rankings, which gives further indications that Kayak is engaging in some kind of joint profit maximization behavior. So overall, our results indicate that Kayak takes other factors than hotel and sales channel popularity into account when deciding about the rankings. And this, why this is not surprising because it's of course a profit maximizing firm, it raises the question of whether Kayak optimizes its ranking only with respect to its own revenues as it claims, or what that takes into account joint revenues of the integrated firm. And assuming low volatility of the cost per click revenues across hotels and time, one could interpret our results such that Kayak favors its affiliated sales channels and the ranking decisions. However, if our even if our results would solely be driven by higher payments of affiliated sales channels like booking.com or Agoda, they could still be consistent with Kayak favoring its own affiliates. And in addition, we also do not observe these patterns that we found for Kayak, the hotel rankings of Google hotels, which is not affiliated with any OTA. So my last point, our analysis cannot provide a definite conclusion on what sort of ranking of hotels and channels is socially optimal. However, briefly, I would like to point out two risks. So first, conditioning product promotions on prices elsewhere can have similar effects as price parity clauses, which have been forbidden in many countries as they could lead to high commission rates and high final prices. And then second, deviations from a best match ranking can harm search quality, of course. However, this regard not clear if integrated platforms achieve worse results in terms of welfare than non integrated platforms. And this is a point of our stop now. Thank you for your attention. I'm looking forward for your comments. Thank you very much, Uli. The next speaker is Tiffany. And I posted some questions in the chat that people asked me. I put them, so Uli, maybe you want to take them. Tiffany. Okay, yeah, thanks for including our paper into the program. And I'm Tiffany Tide from National University of Singapore. And this is a joint work with my colleague Nan Chen. So algorithm steering are becoming an important tool for information intermediation for many platforms. And for example, 80% of the movie watch on Netflix are coming from recommendations. So algorithmic recommendations could have a large impact in terms of social welfare. And these recommendations are generated based on big data machine learning method, but the objective behind the algorithm are set by human. And at the same time, we see these large internet platforms develop vertical market structure. So they are not only the information intermediary, but also players in the related market. So this dual role could create potential conflicts between their profit maximising incentive and how they use the tools for information intermediation. So we already saw example from the previous paper and another example would be that Google were accused for its search ranking bias. That potential favor its own affiliation. And our empirical context is Amazon.com, which is the Amazon marketplace in the United States. And Amazon has a dual role. So it owns the platform, but also it is one of the sellers in some of the product markets on the platform. And this dual role has raised policy attention on whether it is fair to third party sellers and consumers. So in this paper, we would like to examine empirically how this dual role may affect Amazon's recommendations. And specifically whether Amazon is more likely to recommend the product which Amazon is one of the sellers. And in other words, whether there's a steering, and the steering in our paper means that whether the product recommendations deviate from the ones a consumer preferred the most. And we focus on the outside recommendation called Frankly Bought Together. So I'll call you FBT from now on. And as you can see on the slides, the focal product can recommend up to two products to consumers. And the recommendations are made on the product level. So often when we talk about steering on Amazon, many people will think about the design of a buy box. So whether Amazon as a seller is more likely to win the buy box. So we can see buy box as a steering or recommendation on the seller level. But due to our research design, we will focus on product recommendation, which is the steering on product level in this paper. And we collect data in high frequency. And our data cover a six million products that have more than 100 customer reviews at the time of data collection. And we will collect information on prices, sales range, which are used to approximate for sales, the FBT associated with the product, and whether Amazon is currently selling the product. And our research design mainly utilize variation in Amazon's temporary absence due to a Stockholm event. So Amazon product will refer to product whether Amazon is one of the sellers. So when Amazon is out of stock, this means that the same product is still available on the same page. But it is sold by third party only now. And in this case, Amazon's private level product like Amazon basis do not belong to this category because third party sellers do not sell those products. So using the variation, we controlling for real time prices and sales. And we found that when Amazon is out of stock in the recipient product market, the recommendation received of that product from the same referring product decreases. So this suggests steering, meaning that condition on the same pair of products, prices and sales. The likelihood of getting a recommendation depends on whether Amazon is a seller or not. And we also conduct more checks to make sure that the result is not driven by prices or consumer preference related to Amazon's presence. And another variation we use in this paper is the variation in product recommendation in order to estimate how effective the recommendations are. So over time, the focal product may recommend different recipient product. We then use the change in correlation of the sales between referring product and recipient product to measure whether the recommendation is effective in increasing recipient product sales. So let me summarize our findings and contribution. So this paper provides a novel causal evidence on algorithmic steering using product recommendations for large and dominant platforms. And we first show that Amazon products are more likely to receive recommendation than product only sold by third party sellers. However, this does not directly suggest steering because they could be other product features that make Amazon product receive more recommendations. For example, Amazon indeed more likely to sell a popular product or popular or product that are more complementary to others. So to show steering, we use within product variation and we found that when Amazon is out of stock controlling for prices and sales, the same product sold by third party seller is 8% less likely to be recommended. And we also provide evidence that the steering is consistent with platforms profit maximizing incentives. So we found that there's a higher degree of steering in the product category where recommendation can generate more sales. And we also find that recommending product that only sold by third party sellers can generate more sales than recommending Amazon products. This implies that the platform does not allocate the recommendation that maximizes the total sales. And the steering could potentially be harmful for consumers and third party sellers because a consumer did not get their most preferred recommendations and third party sellers product was still away. And given that algorithm is always open about black box, our empirical findings suggest that more discussions and attention on competition policy algorithm make accountability and regulation for dominant platform things necessary. And that's all from me today. And thanks for listening and please let us know if there's any comments and suggestion. Thank you very much, Tiffany, for being on time also. Now, what I'd like to do, we'll have a very short break in the presentation and I'd like to come back to Greg's questions, okay, the big pictures questions, or you can also see that some people have asked some questions in the chat. So Uli, you've had a bit of time maybe to look at the chat or to think about the Greg's questions. Is there one question that you'd like to answer among these? Of Greg's questions. Greg or any of the questions that were asked in the chat. I mean, we can't answer everything but. Okay. Maybe I first start with the chat. So, just one question. Okay. Okay, maybe regarding your C's question. So we got to whether kayak is just reacting to other meta search platforms steering or discriminating against booking.com. So my answer would be twofold. So for a platform that is not integrated, which is Google hotels, we don't see that they discriminate against booking.com. It would be interesting to see whether other integrated meta search platforms like the case of Trivago and Expedia, whether there we observe the same pattern, and then it could well be that these two pairs of platforms just react to each other that each competitor meta search platform is favouring only the only affiliates, and therefore to make up for these loss in demand, they react. But for this, unfortunately, we don't have the data ready, but I think it would be certainly interesting to see that. Thank you. Tiffany is there. I mean, you've had less time to look at the questions in the chat. Right, but I think so for my setting, I think the FBT are the same for all consumers. So this is, so Amazon has a algorithm called item to item collaboration or something. Right, so that means that conditional on the product, the focal product, they send the same recommendation to all consumers. Right, so it's not a personalised recommendation. And the 8% I found is relative to when Amazon is in the market. So the 8% is the difference of likely who are getting the recommendation when Amazon is in the market versus relative to when Amazon is not in the market. Okay, thanks. So we'll have hopefully a bit more time at the end for very general questions. Now I want to move on to the next speaker who is Eslem. Eslem, can you share your screen? Yes, happy to do that. Can you see it? Yeah. Is it okay? It looks fine. Yeah, good. Thank you very much for the very nice introduction and also for having our paper in the program. And I apologise for the weird noise that you hear background. So there's some construction work going on. And that just happens to be at the same time as my presentation. So this is a paper joint work with Simon Anderson, and we study hybrid platform models. So by now you have already heard what it is. I just wanted to emphasise again one fact that Amazon is a dominant player in the e-commerce markets of US and Europe. And Amazon has this hybrid or dual business model. It's a marketplace at the same time and reseller of its own products. And there have been already this policy concerns about steering and other issues discussed previously, so I'm not going to repeat. But I just want to also emphasise that there is also another ongoing debate about the taxation policy of digital platforms. And very recently French government introduced 3% tax over Amazon marketplace revenue and Amazon passed on this tax one for one on its seller fees. So this is another aspect that we will try to also study using our framework. So what we are doing in the paper is basically we try to provide tractable and micro funded model to study trade platforms. And then use this model to understand how this dual role of the dominant platform affects consumer prices, number of differentiated products offered by third party sellers. So this is the variety and the platform profit and the consumer welfare in equilibrium. So in a sense, coming back to this big picture questions raised by Greg. So we look at kind of a longer equilibrium where we account for the entry of third party products in the model as this will be clearer when I present the model. So we have this model of a trade platform here. This initiates basically mediates interactions between a mass of consumers and a continuum of differentiated sellers. But the platform first also decides whether to sell its own product. In other words, to choose to be a hybrid or not. And if he chooses to be a hybrid, it also sets its own price PA. And it also chooses the commission percentage commission that charges the third party sellers for every transaction on the platform. Third party sellers then decide whether to enter. So it basically join the platform. There's a fixed cost of entry. And there's no other cost that they incur. And then if they enter, they choose their own price. On the other hand, consumers, they choose whether to visit the platform and they face intrinsic cost of visiting the platform. And once on the platform, they choose between different products. So if there's a product available by the platform, they can buy product A platform product. That has a different quality VA than the third party sellers product B. And they pay the price of the platform product. And they also get this match value, which is assumed to be random in the model. And if they buy the fringe product, they get this other utility. But they can also walk away buying nothing on the platform if they are outside options better, which also has a random utility. And what we do in the model is we use this mixed oligopoly framework used by the literature basically to study this asymmetry between a big platform versus small or continuum number of fringe firms. So we assume that the fringe firms are infinitesimal so they cannot affect the equilibrium. However, platform product has a mass. So its price is going to affect the equilibrium of the model of the equilibrium demand of the model. This is first aspect. The second aspect we use is this we rely on the properties of longer aggregated games because we model this entry of the fringe firms. And that will give us very nice tractability in the model. And we use this basically the consumers learn their match values to the products on the platform after incurring their visit cost. So we adopt the Weismann search rule basically to study also this two sided network effects. In other words, again, coming back to Greg's question, what would happen if platform most accounts for the number of consumers coming to platform will depend on the number of fringe sellers on the platform. And we also look at this version of the model. Okay. And but we assume is that the fringe firms have sufficiently low cost so that the platform can charge at least some commission to them in equilibrium. And also we assume that this match values are IID distributed with the Gumball distribution, and we look for sub game perfect Nash equilibrium of this game. Okay. So this is all I am going to present about the model. And let me just briefly describe the main results. The first point to highlight. Forget about the equations is that the fringe firms prices just depend on their own costs and incorporating the tax that they pay at Valoran tax that they pay to the platform. And standard logic market, but doesn't depend on the platform products price. And this is also true for the hybrid platform model. Okay, so it doesn't depend on the business model. Why it is the case because fringe firms are small so they are miniscule and they can't affect the equilibrium. And the zero profit condition pins down the equilibrium aggregate. Basically just depending on again the commission charged by the platform and the fringe product cost. So it doesn't depend on the platform product quality or the price. This is another property very important in the model that gives us our main result, which is basically when the platform is a hybrid platform, a better platform product in other words, the higher va quality or lower cost of the platform product increases the equilibrium fee t charged to the third party sellers. And that will lower variety number of sellers coming to the platform, and that will lower consumer welfare, which is surprising because what we are saying is that better platform product lowers consumer welfare, which is not the result that we used to see in standard IO model. So why is it the case that the good platform product is bad news. Indeed, it is because of the fact that the platform has the hybrid business model. So when platform product quality goes up, it lowers number of fringe sellers coming to the platform. So lack of variety fully neutralizes the improved quality of the platform product on the aggregate. So aggregate does not depend on the quality of the platform product. However, the share of the platform product over the aggregate is going to be higher. So that implies that the platform puts more weight in front of its revenues coming from the reseller channel. And that will lead to higher commission on third parties and lower consumer welfare, because in this model consumer welfare is tracked by the aggregate and aggregate is a decreasing function of the commission. So then we look at the platform's choice between when the platform prefers to be pure marketplace versus hybrid. What we show is that there is a cut of quality for the platform product we had in this picture below which platform prefers to be a pure marketplace so it doesn't have sufficiently high quality of the print on product above which it prefers to be hybrid. And this cut off depends on the fixed cost of the platform product. So if fixed cost is nearly zero, platform always prefers to be hybrid because there's always gains from more variety and it can better control its own product prices. But however, when platform product platform chooses to be hybrid, that will be bad news for consumers because now that the aggregate is going to be going down with the quality of the platform product. So that means that depending hybrid platform mode in the market benefits consumers. So this is our critical policy implication that the hybrid platform model in this framework benefits basically banning its benefits consumers by increasing variety number of French products and lowering prices. Coming back to steering concerns if we consider the steering in the sense that platform increases the perceived quality of its own product and lowers perceived quality of the third party products. This type of steering will be only profitable in our framework if the platform product is good enough. But this is the case basically where this will also be likely to harm consumers. And if we look at the other tech issue that I mentioned at the beginning, the taxation, what we look at is that if we tax the platform marketplace revenue in our model of the hybrid platform, then the platform is going to increase its commission to the third parties and that's going to harm consumers. Like basically what we have seen happening in French market of Amazon. So Amazon raised the commissions on third parties in French market, and that's going to be basically punishing consumers, French consumers. However, a unit tax on the platform's own reseller product is going to increase its cost, basically tax is going to increase the perceived cost of the platform product. And that's going to lower commission charge third parties and benefit consumers by increasing variety and lowering prices. So this is, these are the key implications of our framework. And I just wanted to conclude by saying that we provide a model hopefully be tractable also be used by the empirical work to study all this interesting policy questions, because it is really tractable and we use tractability to study how the hybrid mode affects the equilibrium variety prices and we show that the bending hybrid is beneficial to the consumers. Thank you. No, no, I'm done. Thank you. Very nice talk. Next speaker, Leonardo. I'm showing this slide. In the meantime, there are some questions in the chat. Okay, this is, do you see my slides? Okay, perfect. Thanks a lot for having me. And also to Greg for the introduction. This is a joint work with Axel Godi and Shiva Shakar. The paper is very preliminary so any comment is more than welcome at this stage. And needless to say that this paper is another paper on platform duality, which is a very hot topic at the moment. And we try to study when, whether, and how a gatekeeping platform prefers to enter in competition with third parties. We would like to study whether this is good for the consumers and finally whether we should prohibit entry and if so under which conditions. What are the differences between our model and other papers that have been presented in this conference and other papers that are available in general is that we focus on a specific type of platform. So we focus on a digital distribution platform like Apple Store. We introduce network externalities. We have in mind a scenario in which third parties that are hosted by the platform provide the network good. And for instance, if you are a Spotify and you are a consumer, you care about how many other users are using Spotify because you can enjoy user generated playlists. Or if this is a gaming platform, a gaming service, then consumers can play with the games that other consumers. And finally, we look at a specific type of entry that can occur in the market with the product bundle. Basically recently, Apple decided to enter in the market with Apple One that includes music, TV games and clouding services. So let me go through the key ingredients of our model and let us first focus on the case in which the platform stays passive. Basically, the platform is a mere intermediary. And it is charging third parties in Advalor and Commission fee. This is beta. Third parties are monopolies in their own market and they are providing services that are completely different. Third party one can be, for example, Spotify offering a streaming service. And third party two can be, for example, a gaming service. There are three types of consumers. Some consumers only care about music and therefore either they join third party one or they stay out of the market. There are some other consumers that only care about gaming services and therefore either they buy from third party two or they stay out of the market. And there are some other consumers that can potentially buy from both third parties. These are the multi-use consumers. In each segment, consumers care about the intrinsic value provided by the third party. They pay a price P1 or P2, but they also care about the total demand at each third party. These are the network externalities, the direct network externalities that we are introducing in this paper. So what are the key effects in this case? What we find is that although the third parties are not in a direct competition with one another because of the presence of multi-use consumers, there are some product complementities. And this give rise to the typical cwrnw effect. This is a source of inefficiency because prices become too high, becomes higher than the monopoly level. And these prices are exacerbated in the network externalities. In the paper we provide information on how the network externalities are exacerbating this cwrnw effect. But basically the idea is that third parties would like to reduce jointly the price, but in the end they are not able to do so because by reducing price there is an incentive because of the presence of multi-use consumers of another third party to raise the price. This source of inefficiency can eventually be removed by the platform by entering the market. And we show that the platform always prefers in this kind of market to enter the market with a product bundle and to segment the market. Before looking at the case in which the platform decides to enter the market with a product bundle, we need to understand why there is an incentive to do so. Let us suppose that the gatekeeper offers a service that is identical to the one of the third party, a standalone service. If this is the case, the gatekeeper needs to set a very low price, the third party will do the same, competition will kick in, and in the end the gatekeeper will make a little, if not nothing, from the sale of his own product. The third party will make the same, so no profits, and the gatekeeper, given that it is collecting an ad valorem fee, will not make revenue. The only way for the gatekeeper to enter the market is eventually to enter the market with a product bundle, because by entering with a standalone product will not make any revenue, so therefore in the first stage of the game will have no incentive to enter the market. The gatekeeper can enter the market by segmenting the market, by segmenting the consumers in the different segments, and in order to do so it needs to act as a price leader by setting a price PA such that there is no incentive for the third parties to deviate from the monopoly profits, the monopoly prices. What we show in the paper is that there exists an equilibrium in which in the end third parties are setting a monopoly price, the gatekeeper is setting a price PA, and this price PA decreases in the network externalities. In other words, in order for the gatekeeper to enter the market, the gatekeeper needs to set a sufficiently low price, and this price is reduced when network externalities are sufficiently large. Now, the question is whether given the two scenarios that we have highlighted, there is an incentive for the gatekeeper to enter the market. Now, what we find is that there exists a critical value of the Advalor and Commission fee, such that if the commission fee is sufficiently low, then the gatekeeper prefers to enter the market. And the reason is that it can open a new channel and extract surplus from the consumers. On the other hand, if the Advalor and Commission fee is sufficiently large, there is no point for the gatekeeper to enter the market. But what is critical here is that this commission fee, the critical value of the commission fee decreases in the network externalities. And the reason is that if the gatekeeper is entering the market and network externalities are getting sufficiently large, then the gatekeeper needs to price very aggressively in order to induce the third parties to segment the market. This reduces profits compared to a situation in which the gatekeeper can be passive, and therefore entry will occur less often when network externalities are sufficiently large. Now, the second question is whether the gatekeeper entry can benefit consumers. And here we have two main effects at stake. On the one hand, the end of the gatekeeper in the market will reduce prices. It will remove the product complementarity issue, and therefore the Cournot effect, prices will go down and eliminate a source of inefficiency. This will have a pro-competitive effect. On the other hand, however, the entry of the gatekeeper will fragment the consumer demand. The single-use consumers will not be able to benefit from the interaction with the multi-use consumers. And therefore, if network externalities are sufficiently large and the gatekeeper is entering the market, this fragmentation of the demand will outweith any pro-competitive effect, and therefore consumers will be worse off even though they are paying a lower price, given the fact that the pro-competitive effect of the gatekeeper entry. Now, in the paper, we finally discussed some policy implications that are related, for example, to the regulation of the commission fee. By regulating the commission fee and prohibiting the gatekeeper to set an independent fee, we put a pressure on the Advaloran fee stimulating more entry. And what we show in the paper is that entry will not necessarily be a good news. And we also discussed some policy issues that relate, for example, to abandon duality that has been proposed in the US. And what we show is that abandon duality may ban innocent in pro-competitive entry. Therefore, it might create some unintended effects. Finally, in the paper, we also discussed some potential remedy that can arise, for example, from the interoperability of services. And this might mitigate the fragmentation of the network effect that the entry of the gatekeeper in the platform might create. That's it for today. Thanks a lot for your attention. If you have any questions, please get in touch with us. Yes, or use the chat. Next, so we were running a bit tight on time. So what I'd like to do is get over with the presentations and then we can have five minutes remaining some more general questions. Who's next? I think it's Wynn. Okay. So hi, everyone. I'm Wynn and I'm going to speak about the topic on data usage and strategic pricing. This is joint work with Xin Yi Liu. So we've already seen a lot of excellent background information about the Amazon case from previous speakers and how Amazon's entry into the product market might hurt independent traders. So but what is missing from this, the literature, its first data. So as for Staya emphasise in her speech. So the question that lies at the heart of the Amazon case is data. It's not only about Amazon entering or not entering, rather it's about data. So whether Amazon's access or use of this data may give it an advantage or would this marginalize third party sellers. And also as more explicitly considered in the US, although there is no formal case yet. But the Wall Street Journal investigation found that Amazon using data from third party sellers to develop its own product actually violates its own companies policy, which says that it will use category level data instead of individual level data. So there is this second missing ingredient in the literature, which is the extent of data usage. So we are going to incorporate these ingredients into our model and try to investigate how independent sellers would respond to potential platform entry. So the emphasis here it's about the Exxon impact of data usage on competition. And in addition, we consider different extents of data usage, which ties us back to the case in the US, and we consider three forms of entry. So the first one is random entry. So the platform in this case doesn't use any sales information and enters with a randomly selected product. Secondly, we consider category entry. In this case, the platform uses category level data and enters with a product in the best selling category. Under targeted entry finally, the platform uses individual level data and enter with its own version of the best selling item. So the main result is that more detailed information usage by the platform softens competition between independent sellers. So let us try to compare random entry to category entry first. So on the random entry price and sales in the early period have no effect on platform entry. But then under category entry, they have an impact. So this gives sellers incentives to raise price and limit sales to deter platform entry. So more subtly, let us compare now targeted entry to category entry. So a seller can be strong or weak. So for a strong seller, it has strong incentives to raise price and that's because A, it sells fully determined the entry strategy of the platform under targeted entry instead of partially under category entry. And B, it would be replaced for sure under targeted entry. So together, this implies strong incentives to raise price. Now for weak seller, it wouldn't be replaced under targeted entry. So it has weaker incentives to raise price. But then a lower price in this case actually reduces the chance of platform entry because it's going to reduce the sales of the strong seller. So all together, whether the weak seller has stronger or weaker incentives to raise price, that depends on the fee structure. So we consider both per unit fee and proportional fee. So under per unit fee platform entry, it's going to hurt the weak seller. So it is going to lower price to deter entry. But under proportional fee, the weak seller benefits from platform entry. So it has incentives to raise price to attract entry. In summary, the effects on the strong seller dominate, so prices are higher under targeted entry. Okay, so now a full-flatch welfare analysis of all three extents of entry can be found on the paper posted on the conference website. If you would like to go deeper, you are very welcome to read that. But in this presentation, I'm going to highlight the comparison between category entry and targeted entry because this is probably the most relevant scenario for the Amazon case. So more data usage, whether it benefits sellers, platform, consumer, it's going to depend on the degree of competition between sellers so that could be high or low. Also, the fee structure and fee level. Okay, so we note on this table, there are cases where sellers can benefit from a move from category entry to targeted entry. So a more intensive usage of data. And especially that's when the degree of competition between sellers, it's high. So then the benefit from softening competition is the highest. Okay, so now on the proportional fee, we also note that the platform can benefit because it has a more aligned interest with the seller. And probably this is the fee structure that Amazon is adopting. So we highlight this case. And interestingly, we also note that there are cases where consumers can benefit from more intensive data usage. And that's when the competition, the degree of competition between sellers, it's low and per unit fee is adopted. So that's because they can benefit from the elimination of double marginalisation. And interestingly, this is actually a case for data usage from the consumer's perspective. Now, in closing, we have asked the question of whether a platform can encroach the market of independent sellers, given its information advantage. And we emphasised different extents of data usage by the platform. So the main result here is that whilst data usage can hurt independent sellers exposed, it can relax competition between the mix zone. And the reason is that data usage gives sellers incentives to manipulate market signals, and it could actually benefit both sellers and the platform, although consumers may be hurt. So more generally, in this paper highlights the importance of market dynamics. So if there's a huge discussion saying that how a seller could be hurt by a platform, if I were a seller, I wouldn't say put. So I think it's equally important for the regulator to consider Exxon incentives in addition to exposed incentives. So thanks very much for listening. I'm going to pass the mic to Joe and take question afterwards. Thank you. Thank you, Wynn. Indeed, Joe, it's your turn. Thanks, Alexandra. Let's see if I can share this with everyone seeing this. Anyway, so I thank you very much. So I'm Joe Perkins. This is joint work with Jorge Padilla and Salvatore Piccolo. I should flag up that this is part funded by the Music Streaming Service Spotify. All views are of course our own. And I'm going to talk about self-referencing by gatekeeper platforms. I'll talk a bit first about the motivation. Why did we start looking at this? And then I'll give a very brief sketch of the model and some of the main results therein and maybe talk a tiny bit about the implications as well. So overall context where you've had a variety of complaints most particularly played out at the European level against the behaviour of dominant gatekeeper platforms or alleged gatekeeper platforms in the mobile phone market. So that means essentially Apple and Google Android. And some nice, I suppose, marketing campaigns on some of these things about the alleged kind of turn of companies like Apple towards being dominant firms. But there is some serious and interesting economics I think in these issues. So, you know, what we can think about here is you can think about a smartphone market with an aftermarket, a couple of dominant firms in the smartphone market, a range of, in this case, music streaming services. Spotify and Deezer, there's actually others, Bandcamp is highly recommended for instance, but with some of those music or one each of those music streaming services controlled by the platform, so Apple Music or YouTube. And so the allegations are that self-preferencing is going on and furthermore self-preferencing between the smartphone and its music streaming service and furthermore that self-preferencing is in some ways harming consumers. What kind of self-preferencing are we talking about here? Well really we're talking about a combination of raising rivals costs. So imposing what are alleged to be excessive fees for services like Spotify or Epic Games to sell to consumers and also tying in with in-app purchase. So requiring that purchases within an app have to go through Apple's mechanism or have to go through Google Android, Google Play Store. So that's the, I suppose, the background there, but we wanted to see how this would work in a theory model and whether indeed, I think related to Paul and Diane's questions on the chat, whether indeed they seem to be rational, perhaps for closure or excluding third parties, was in fact rational behaviour on the part of the dominant firms. So very brief sketch of the model. We've got a two-period model with monopolist provider of devices and Bertrand competition among homogeneous providers of homogeneous services. It's obviously a simplifying assumption. Marginal costs are normalized to zero. You can think about a wide range of applications here, music, video games, dating services, and others. So demand for the device changing over time, which I'll talk about a bit more later. And what happens in this multi-period model is that at the start of the second period, the monopolist can invest to foreclose in the linked services market. That eliminates competitors and replaces the services provided by a third party with its own product. We assume that the monopolist product is weekly inferior. That's the impact on results as we talk about later. Now, that need not literally be an alternative product. And you might also say, well, we're fully efficient bargaining. The monopolist could extract surplus by contracting with one of the third parties, which is right, but we see that as pretty unrealistic in practice in this setting. So this is the game structure fairly straightforward where you have the monopolist first selling devices at the start of period one on the left. We assume of choosing whether to buy those devices, the third party suppliers, then selling the linked apps. And after that, the monopolist can then choose whether to foreclose the app market, the services market. And then again, you have either a repeat of the first period game if it doesn't foreclose, or you have a slight tweak of the first period game in the sense of the monopolist is then setting its prices in the linked services market, the app market. So that's a very brief sketch of the model, obviously much more detail in the paper. A very brief sketch of the results as well. What we find is that there exists equilibrium with foreclosure. And there also exists equilibrium without foreclosure. And these are unique depending upon parameter values. Essentially what you find, and this is shown in the charts at the bottom. Lower values, you have a foreclosure equilibrium, so towards the x axis you have a foreclosure equilibrium further away. You don't have a foreclosure equilibrium. What you see is that where beta is high, so the quality of the monopolist app is fairly close to that of the parties. And when demand is not growing very quickly. So this is the x axis, which is the theta value. Well, in those cases, you're more likely to see foreclosure. So slow demand growth and a relatively weak, relatively strong product of the monopolist, you're more likely to see a foreclosure equilibrium. And what are the welfare results we get out of this? Well, again, I think this does relate to Paul's question about potential irrationality. Here we have a timing consistency problem that the monopolist would want to commit not to foreclose ex ante is profits are lower with foreclosure than without foreclosure. But when it comes to the second period, it often wants to foreclose anyway. And I wouldn't place too much emphasis on this, but there is a statement by Steve Jobs back in 2008 saying we don't intend to make money off the app store. That's could be seen as a commitment or an attempt to commit not to foreclose, I guess. So that's the position for the firms. Finally, I suppose what really matters from a policy perspective, what happens to consumers, perhaps not surprisingly, consumer surplus is lower with foreclosure than without foreclosure, assuming that the quality of the app provided by the monopolist firm is lower, is of lower quality. And that consumer harm, again, is larger if market growth is relatively slow. And also if the quality of the third party, the quality of the monopolist app is relatively poor. So, for a brief discussion, because I think we are almost at that time and Fred, foreclosure has two types of effects for the monopolist. It reduces the demand for the device, but it also enables the monopolist to exploit its install into customer base. So you have one effect which is trying to grow the ecosystem, but another effect which is exploiting that install customer base, and the balance between those depends upon a range of factors such as a cost of foreclosure and the level of demand growth. I've got one or two just empirics, which I think at least don't contradict this kind of model, but I won't go into those in any depth. To give overall conclusions here, I think to be honest, this is a model that could be applied to physical markets too. I think digital markets can pose similar problems to physical ones, but often in new ways, and that ability and incentive to lock in and to extend monopoly power is potentially greater in digital markets. I think Greg was talking about policy implications. There is, I think, a case for a special responsibility on those gatekeepers where they compete with third parties, as in, for instance, music streaming market, and also that the scope for consumer lock-in might support more active intervention to enforce interoperability, or more radically perhaps opening up these kind of closed ecosystems like the Apple App Store or Google Play. Okay, thank you very much.