 So welcome to the Parallel Session number three, competition on and between platforms. So Yossi Spiegel is going to introduce the session. I don't think I have much to say about Yossi, he is a super well-known economist, a professor at the College School of Management at Tel Aviv University. So Yossi, you have 15 minutes. Thank you so much. Okay. Thanks. So let me share my screen. Can you see it? Okay. So this session is about platform competition on platforms and between platforms. And it has six papers that deal with issues ranging from pricing to platform design to platform management to empirical for theoretical. What I found interesting is that I went and checked today what was done on the first conference that was held in 2001. And so you can see that at the time there was another pricing issue, but I guess it's all gone now because we don't speak much about internet connections and often on net pricing. But interestingly, still people are worried about obfuscation, which is something that we are going to hear today as well. And this is the classic paper of Kayo and Julian about two-sided markets and pricing in this. And what is interesting is that at the time in 2001, the Microsoft case was still drawing a lot of attention and research. And I guess that soon we are going to all talk about the Google case and the Facebook case and all of these other antitrust cases. So let me give you a short preview of the papers because the presentations are very short. So this is my take on what the papers are doing. And then, obviously, there are lots of details to be fit in by the authors. So the paper by Doshin and Patrick deals with platform pricing. And the question that they tackle is why is it that lots of platforms are charging high commission rates? So people complain a lot about the fact that you have to pay Amazon up to 30% commissions and up stores 30%. And hotels complain bitterly about the prices. They have to pay booking.com and restaurants complain about the prices that they pay to vault. And maybe there is going to be some antitrust intervention. And the question is why are these rates so high and what are the welfare consequences? And the main insight that the paper brings about is that somehow competition doesn't work because competing platforms have an incentive to charge very high ad valorem rates, revenue shares. And those lead to fewer apps and lower welfare. So they harm welfare. And I guess the main insight is that the platforms do not internalize the negative effect on the apps of having such high rates. And this is due to multi-homing. And I guess one thing to notice during the presentation is the key issue here is why exactly a platform does not internalize the negative effect of the high fees on the supply of apps. The second paper of Ron Bergman and Yuval Heller is about pricing, not so much about platforms, but still it's about pricing. And the question that they ask is why firms often underestimate demand elasticities but overestimate advertising elasticities. And the main insight is that this suboptimal behavior confers strategic advantage on firms vis-à-vis rivals. And so given that the conference is in to lose, it's worthwhile noting that firms underestimate demand elasticities to become puppy dogs instead high prices and induce rivals to do the same. But they overestimated advertising elasticities to become top dogs and gain strategic advantage via rivals that become softer. And I guess one common question that to keep in mind is how exactly the paper, what are the implications of the paper for platforms. One thing to keep in mind is platforms often provide sellers with detailed information about sales because they collect all of this data and they can provide it to sellers. So in this sense, if firms wish to be uninformed about demand elasticities and then this leads to high prices, high prices lead to low volume of sales, platforms like to have high volume because they get commissions and maybe they wish to provide this information. And I guess another question to keep in mind is that if you compete against the very sophisticated rivals, say you sell on Amazon marketplace, but you also compete with Amazon. Amazon knows a lot, it's very informed, so maybe this tactic wouldn't work, but this is something to keep in mind. The third paper by Johan and Shomugi deals with platform design. Here the question is not about pricing per se, but rather how much information should platform provide buyers about seller's fees or does it help a platform to shroud some of the fees because those are common and the question is why is that. So the key insight is that shrouding seller's fees may attract naive buyers to the platform and therefore once you have more buyers on the platform, it's easier to attract sellers and therefore you can raise seller fees and this is where the benefit comes from and I guess one thing to notice when you hear about the paper is to understand exactly why shrouding fees affects the number of consumers because this works in the paper in a subtle way and I guess the exact mechanism is interesting to notice because this is the key to the result here. The fourth paper by Bakus, Blake, Petus and Tadellis asks the question is it a good idea for a platform that connects buyers and sellers to allow them to exchange messages and obviously this is a design feature because you can open a messenger and allow buyers and sellers to communicate and the main result here, the main insight is that this was actually done on eBay in Germany and allowing buyers and sellers to communicate increased the likelihood of successful bargaining by 14%. This is quite substantial showing that communication helps but another insight from the paper is that you know just allowing people to communicate doesn't help immediately. People have to learn how to communicate effectively only once they learn the benefits kick in and one thing I guess to notice about the paper and to keep in mind is that you know what exactly communication does why exactly does it facilitate the bargaining and which type of messages can help you to complete negotiations effectively. The next paper by Frank Schluter looks at platforms most favored the nation's clauses which essentially say that you cannot charge a lower price on your direct channel than the price that you set on platforms and I mean this was a big debate say in Germany where platforms were not allowed to impose MFNs. MFNs like to impose those MFNs because they say that the problem is that you know they provide a lot of information to prospective customers, customers check the website but then they go on and book a hotel on the hotel channel and pay less and therefore there is a showrooming problem. So here the paper offers a new theory of harm that was not offered so far and it says that under MFNs platforms prefer to facilitate sellers collusion. This is kind of surprising because normally we think that the platforms like sellers to compete because it increases the volume of transactions but here actually platforms like sellers to collude and charge high prices and moreover MFNs allow sellers to collude more effectively so both considerations platforms have an incentive to promote collusion and MFNs also facilitate collusion so one concern about MFNs after seeing these papers that facilitate collusion and I guess the key issue here to notice is why exactly MFNs make sellers like collusion. And again the explanation here is a little bit subtle and I guess this is the key thing to note when you listen to this paper and exactly what is the mechanism by which they facilitate collusion. And in the last paper by Liu and Chan asked the question can platforms promote quality of seller services by inducing only some sellers to improve quality. In other words the question is if a platform likes sellers to provide high quality should it incentivize all sellers on the platform and the main insight here this is an empirical paper based on data from eBay is that in fact a platform can only incentivize a fraction of all sellers and other sellers who did not get any incentives will also have an incentive to adopt the higher quality. The higher quality here is premium service faster delivery longer return periods and I guess the insight here is that for a platform to incentivize sellers you don't have to go after all sellers it's enough that you incentivize only some and so this is because there is a spillover and they also find that the spillover is larger when more sellers get incentives and I guess to me it suggests that there might be some strategic complementarity in the adoption of quality if if a rival of yours provides better service you have to provide better service as well otherwise buyers will not buy from you. And I guess one you know the key issues to keep in mind at least in my mind is you know we already know that the platforms use seeding in other words they give incentives to some customers and then somehow these incentives propagate throughout the system and here it's kind of seeding on the supplier side rather than the buyer side so I guess you know it would be interesting to know what are the relationships between the two. And also one thing that comes to mind is input output framework where you where where if you if you if you have a complete network it's enough that you touch only parts of the network and then things propagate throughout the network on their own and I guess another question is how effective the mechanism is they show that this exists but the question is you know how exact how effective it is and okay so I guess we have very interesting six papers to listen to that advance the literature on platform competition and I'm looking forward to the to the presentations. Okay so thank you so much Yossi for this great introduction so I'm happy to give the floor to the speakers so I would just like to remind people to send the questions by chat so each speaker has eight minutes and after each speaker will have time for a couple of questions so if you'd like to send the question by chat to me or to everyone I can read it loudly and and so that we can engage in some discussion. Okay good so so the the first paper of the session will be presented by Matthew Bakus from Columbia trying to work with Tom Thomas Blake, Jeff Petus and Steven Tabellis so I'd like to give the word to Matthew. Are my slides showing? Everybody can hear me? Perfect thank you thank you to Yossi for that introduction Renato for running this the the organizers for this opportunity to share the work and my excellent co-authors Tom Jett and Steve this is a paper that as Yossi said will examine the effects of the availability of a communication technology on the likelihood that parties to a negotiation successfully transact and I'll spend just a moment trying to to motivate why we're interested in that question there's a lofty argument that as economists we should care about bargaining protocols and how they affect the likelihood of success whether in an online platform like the one we study peace negotiations climate change mitigation and here I've given some examples of bargaining technologies that you could think of as ways of moderating communication in the bargaining process that's quite lofty of course and fodder for an introduction so perhaps the more direct case for why we should care about this is related to to our actual empirical setting that platforms that are managing buyers and sellers have to make decisions about whether to allow buyers and sellers who are negotiating to communicate and here I've listed some examples of platforms such as Tau Bao, eBay, Amazon would be another that have explicit bargaining protocols on which billions of dollars are transacted every year that are making different decisions about how much communication to allow and in what form the main contribution of the paper though excuse me is that you know while there's been a lot of theoretical and experimental work on this question there's been essentially no introduction of evidence from real life bargaining in the field and so this is going to be the first point of evidence we're going to use data from eBay's best offer platform and we have a natural experiment in the in the rollout of the messaging feature there are two main results one that there's an economically and statistically statistically significant positive effect of communication on bargaining success and two inspired by some of the findings digging into that first we show that there are dynamics that you could interpret as evidence of learning that sellers are learning how to use repeat users are learning how to use the communication feature all right let me tell you a little bit about the best offer bargaining environment we've done a lot of work on this typically on eBay.com so here's an example where I've done a search on eBay.com for Jean-Michel Besquiat I found a painting that I liked the one in the middle so these are search results if I click on the listing in the middle then I will go to the view item page where I have two I see more information about the item but I also have two options for purchasing it one would be to add it to my cart and check out later or to buy it now in which case I'll pay four thousand four hundred and twenty five dollars or because the seller has enabled a free feature called best offer an extra button appears and it's right here under add to cart it's make an offer if you click on that then what you'll get is the pop-up at the top in this slide it prompts you to offer a numerical amount for the seller and if you send that offer they will have 48 hours to accept reject or counter if they counter you will have 48 hours to accept reject or counter and so on in a manner that looks a lot like Ruben Steenstahl alternating sequential offers bargaining we're not going to be working on eBay.com for this paper though we're going to be going to the german version of the website which is a functionally separate website with a little bit of an idiosyncratic history what I want to call your attention to is the difference in the make offer pop-ups on the two sites on the one hand one's written in german but on the other there's a missing feature on the german version which I've highlighted in the red rectangle on the bottom there is no option on the ebay.de website to add a message to seller these messages on the US website are limited to 250 characters and they can accompany any offer this was the case at least until 2016 when on may 23rd ebay.de turned on the communication option now there's not a lot happening on ebay ebay.de at the time most importantly there are no other major site changes no other confounds that were rolled out simultaneously also important the website was updated so desktop users are going to be treated in the post period but the mobile app was not and during this period about 50 of ebay.de users were using the mobile app what's also the case is that adoption was almost instantaneous and about six percent of bargaining interactions in the post period involving a buyer using a desktop involved a message on this feature okay so it was a low compliance rate but it very quickly jumped to that level now since this is a brief presentation I'm going to jump right in and show you the headline table of the paper which is here I'll skip columns one through four which are the pre-post estimates and focus on the diff and diff in columns five and six first the dependent variable here is whether a bargaining interaction namely a buyer item pair successfully ends in a transaction we're regressing this on dummies for post and desktop and in the diff and diff the coefficient of interest will be on the interaction between being in the post period and being a desktop user what we find is about a half a percentage point four percent point effect on the likelihood of successful transaction now note that's an intent to treat estimate in other words it's point four percent for the entire platform so when you're evaluating this for economic significance what we'd really like to know is the more interpretable intent sorry treatment effect on the treated in other words for people who actually use this feature what was the effect on the likelihood that they interact and so the common trick to get at this is to convert our diff and diff estimator to an instrumental variables estimator by using the post desktop dummy as an instrument for whether you send a message in the post period as a desktop user and that's what we're doing in column seven and eight and so what that allows us to do is say that we see between a seven and eight percentage point effect for the subset of users remember it was only about six percent of bargaining interactions for the subset who actually take advantage of the feature so we think of this as being economically and statistically significant what I'm going to show you next is just another version of that diff and diff regression which does weak specific effects now this is often done in the applied micro literature to ask whether the effect proceeds the cause in other words whether we get significant effects before the introduction of the change and we do not but what's more interesting in this figure is that in the post period the effect doesn't take off instantaneously it takes a few weeks to kick in and then it stabilizes around point four percent and that's what we wanted to dig into next we're actually going to use the text of the messages that people are sending to try to understand what's happening in the dynamics of that effect I don't have time to cover this whole slide but I promise it's pretty standard in the text analysis literature we're going to take all the messages that are sent in a particular week convert them into vector space by literally counting how often a particular word appears that's called a bag of words approach then we can use cosine distance since we're in a vector space to measure how much the language that people are using is changing that gets represented in these two heat maps which are telling you about the cosine distance between the set of words that are used in say week one versus week eight and the first thing to notice is that we get nothing in particular on the buyer side but on the seller side it seems like there are some interesting patterns here which turns out to make sense because the modal buyer shows up once and then leaves the platform there's no room for learning but the modal seller is a repeat user now these heat maps are pretty hard to interpret so here is the slide where I'm going to ask you to do a little bit more work but these are an attempt to try to make the heat maps easier to approach in panels a and b we're going to look at the bottom gradient of the heat map which is comparing the week indicated on the x axis to week 10 and asking how different they are again for buyers we see noise and let's focus for now on the solid line but for sellers we see a slow descent which corresponds to the idea that what people are saying from week to week is changing and getting closer and closer to what they're saying in week 10 what the different lines are doing is conditioning on sellers who actually send more messages during the 10 week period so if we condition on sellers who send more than six messages we see a much steeper descent suggesting that what they're doing is changing even more so the first thing we've learned is that these patterns are related to repeat users both in the difference between buyers and sellers and the difference between low repeat sellers and high repeat sellers so the next thing we wanted to dive into is to ask if this is learning then we should see that convex pattern that's the signature of learning and information models does the rate of change slow down as time passes for that we plot the off diagonals of that heat map that's telling us about the rate of change again for buyers we see noise but for sellers we see that the rate of change is slowing down over any level of differencing here it's the one period difference two period difference three period difference and so on so we take this to suggest that we're seeing the convex pattern that we would take as a signature of learning the final result I want to show you is sort of the cherry on top if you thought that they were learning one would hope that they were learning something that would be correlated with success and so what we're going to do is look at outcomes for these messages so this is a regression at the message level that regresses a dummy for whether an offer is accepted on the similarity of that message to the corpus of week 10 messages in other words what we're converging to and across a bunch of different specifications controlling for message length controlling for seller fixed effects we get a positive effect note that we lose statistical significance when we control for seller fixed effects there's just a lot of them what's also neat here though not only does this suggest that if you say things that make you look more like an experienced user you're more likely to transact we also see that the sizes of the effects are comparable to the size of the effect that we saw for the treatment effect on the treated which was between seven and eight percent okay I think that's about all I have time for so let me go ahead and wrap up we asked whether communication facilitates successful bargaining we found that it does that it has a large effect and we also were interested in sort of this aspect of identifying learning behavior in this in this communication equilibrium of course our paper raises more questions than it answers what a bargain are saying that effects outcomes what are the mechanisms what's the best communication protocol we do reflect on these problems we have some descriptive evidence in the in the appendix we can talk a little bit about more more experienced sellers are saying but this evidence is entirely descriptive we don't have causal variations that would allow us to identify that so we hope that there will be more work on these questions thank you thank you thank you so much Matthew for this this great talk I would have liked to have questions now but since we're out of time I would like to postpone questions on on Matthew's presentation for the end to make sure we don't we don't explode the time so so now I would like to give the the floor to dosing Jean who's gonna present joint work with Patrick Ray and again you have eight minutes and questions if anyone wants to ask questions please please do it by the chat can you hear me okay so I'm very happy to present this joint paper with Patrick Ray the paper is yet very preliminary this is about App Store I cannot okay so the 30% commission charged by Apple's App Store and Google's Play Store has received the hesitation both from media and policy makers and one argument against the high commission might consist and it's a negative impact on innovation and a possible defense of Apple and Google may consist in arguing that competition between iPhones and Android phones induce them to promote app development so here we study how competition between two smartphone platforms affects their app store commissions and thereby app development we consider two scenarios for comparison so app store competition can be done in terms of ad valorem rates or it can be done in terms of wholesale prices and if we consider as a benchmark of the static situation in which there is a given number of apps already developed then ad valorem rates generate more efficient outcome than wholesale prices because on the ad valorem rates there is no double marginalization however if you look at the main scenario of dynamic situation in which there should be endogenous development of apps then the result is reversed and ad valorem rates generate innovation chalk up meaning that there is no app developed while wholesale price some positive amount of innovation our model mobile platform competition is such that each mobile platform is vertically integrated meaning that they sell smartphones also they operate their own app stores so there is competition on both sides on the consumer side by choosing price of smartphones and on the app side by choosing app store tariffs and we assume consumer is a single home but developers are multi-home and we use hotel model for the device market so the timing is that first they compete on the app side so its platform choose the the tariff of its app store and then innovation decision of each app developer is done and and and then the competition on the consumer side unfolds so if you consider static situation so let's look first what happens if the platform choose wholesale price and then an app developer's profit is given by this where p minus w i is the profit margin so here app developer is going to choose monopoly price based on marginal cost w i but if a platform choose ad valorem rate then the profit of an app developer is given like this so as long as this one minus a is positive the app price chosen will be just the monopoly price based on general marginal cost so there is a notable marginalization and on top this monopoly price does not depend on ai so we can conclude that in a static situation ad valorem rates generates a lower price than any positive wholesale price now in general the tariff chosen by an app store determines the pie and its division by the pie we mean the welfare generated by an app expected welfare generated by an app per consumer and so this circle is the pie and this pie is divided into consumer surplus platform revenue and developer profit of course all this division is done by the choice of a tariff wholesale price or ad valorem rate now if we compute and solve for everything up to the first stage where it's from choose its app stores tariff we have this expression for its platform's profit where t is the transportation cost parameter and y is the number of apps so given number of apps essentially each firm is competing in terms of the sum of consumer surplus and this is a platform revenue per app okay so if we consider the benchmark the static situation where there is a given number apps already developed then its platform just maximize the sum of consumer surplus and platform revenue and this leads to choose ad valorem rate equal to maximum equal to one if competition on app store occurs in terms of ad valorem rates but if they compete in wholesale price they choose some positive wholesale price but not that extreme but anyway so in static situation since there is no double imagination and ad valorem rates welfare is higher with ad valorem rates than on the wholesale prices so now let me give you the intuition about why a equal to equal to one and ad valorem rates because the price is constant it does not depend on ad valorem rate it means that the pie is constant even if a platform increases its ad valorem rate on top consumer surplus is also constant since the platform maximized the sum of platform revenue and consumer surplus by increasing ai the platform can increase platform revenue and reduce develop profit this is why its platform ends up choosing a equal to one but things are different if they compete in wholesale price because increasing wholesale price exacerbates double imagination so the pie is smaller consumer surplus is smaller so the platform does not want to charge excessively high wholesale price so finally what happens in the dynamic situation where apps must be endogenous developed then actually we can show that there is a unique equilibrium which is symmetric and this so its platform still maximizes the same objective but the reason is that in symmetric equilibrium you get the faster the condition the impact through endogenous y in in the faster the condition is disappears because the in symmetric equilibrium this term is a zero so that means that still it's pumped to the same tariffs for app stores implying that on the other valorem rates there is innovation choke off so now ad valorem rates generate lower welfare than wholesale price so let me conclude um um competition does not provide uh any incentive to promote app development because of a multi-homing of apps so uh so in the future we want to uh uh extend the analysis for instance by considering uh platform specific uh adaptation cost etc so thank you very much for your attention thank you so much Lucy for the great talk i believe we have time for for one question if we do not have a question so far we can have them at the end let's let's see if they're fine um so so so now let's move to to the third presentation the speaker is Meng Yu from Washington University who's going to present joint work with shanghui and touch on from Washington University as well so uh thank you the floor is yours thank you have eight minutes hello everybody my name is monlou and i first want to thank the conference organizers for including our paper into the wonderful conference program also want to thank you see for this wonderful introduction given that our paper is still early stage so it certainly would benefit the comments particularly on the framing of the paper so um this is is a joint work with shanghui i believe is in the audience and also touch chance so we are with the marketing group at Washington University in st louis so motivation of the paper uh monetary incentives are commonly used by uh platform designers to not just set our effort in certain ways for example amazon gives a 75 discount in storage fees to sellers who store their popular products in amazon's warehouses in order to uh you know uh improve the delivery and the inventory process but giving monetary incentives to all sellers on the platform can be costly for platforms so sometimes platforms may want to offer incentives to only a subset of sellers and we call this practice targeted incentives so in this paper we are going to study whether targeted incentive works as well as the underlying mechanisms so our study leverages the policy change on ebay in uh back in 2012 when ebay offered a monetary incentive to enhance seller quality on shipping and return services for example uh specifically sellers who offer a generous return period for their customers and a fast handling of their products would be eligible to receive an additional five percent discount of the commission fees so what we we call this practice this promoted behavior premium service or ps one key feature of this policy is that the incentive is targeted to only ebay top rated sellers or etrs who are certified sellers on ebay who pass a you know pre-specified quality threshold so they are considered of higher quality in the eyes of consumers so this incentive the nature of the incentive that is that it is conditional on the status of the seller as well as the seller's behavior so here we plot seller's quality provision in the 12 weeks before and 12 weeks after the policy change separately for the targeted and non-targeted sellers first we see that the share of ps listings or the quality provision is consistently higher for targeted sellers than for non-targeted sellers even before the implementation of the incentive and this is consistent with the fact that targeted sellers are platform certified sellers who are higher quality sellers to begin with and secondly we find that immediately after the policy change both targeted and non-targeted sellers increase their quality provision the fact that non-targeted sellers also are more likely to exert effort after the policy change even though they are not qualified or not eligible for such such incentive and this suggests that the targeted incentive would have some spillover effect so to understand why we consider four substitutes from the consumer's perspective defined by targeted and non-targeted sellers and by high quality versus low quality product so first we consider the two products by targeted sellers so what the monetary incentive does is essentially reduce the cost of quality provision for high quality products and this would mean an increase of the supply in the high quality market and reduce the supply given that the total supply of targeted sellers in the short run is fixed this in turn leads to a reduced equilibrium price and increase the equilibrium quantity for the high quality product and also an increase the price and reduce the equilibrium quantity for low quality products of targeted sellers so the theory seems to be pretty clear but how do we empirically test this prediction then well just by looking at these graphs it'd be great if we have you know many replicas of the market with different treatment intensities that basically shift the supply curves differently so that we can track the changes in the equilibrium notions and see if they're consistent with predictions and luckily we have that kind of variation in the data specifically we leverage the large number of product markets on ebay and these product markets for example cell phones and smartphones or toys would be standalone markets these markets vary in the ex ante share of etrs sellers or targeted sellers a market with more etrs sellers to begin with would be more treated than the market with fewer etrs sellers so essentially our identification is a continuous DID that compares the temporal changes in outcomes across markets with different shares of targeted sellers and we indeed find evidence that the supply and equilibrium quantity increase for high quality products of high of targeted sellers in more effective markets although the the coefficients are not very significant significantly estimated but we do find a strong evidence that the supply and equilibrium quantity is reduced for the low quality products and the equilibrium prices increases for the low quality products of targeted sellers but how does this affect non-targeted sellers then well in the eyes of consumers the targeted sellers are platform certified sellers who command who can command higher prices for their products in other words they should be the sellers up in the quality spectrum given this it is plausible although the theory can go both ways but it is plausible that high quality products of non-targeted sellers are more substitutable to the product on the right than to the product on the left so if so then the rising price of the target sellers low quality product which is the product on the right would make consumers substitute toward high quality products by non-targeted sellers okay which moves the demand curve outward and this in turn leads to a you know an increase the supply given that sellers anticipate an increase the demand and this in turn leads to a reduced supply of low quality products from non-targeted sellers given the fixed total supply of non-targeted sellers in the short run so to confirm the rising demand in the high quality non-target sellers market we study the ps premium which is the difference in sales probability between identical products that only differ in ps so this requires that we match the listings in several key aspects to control for a product seller and market level heterogeneity so that they are otherwise identical except for ps so I don't have enough time to go into details but let me tell you that we find a strong evidence that consumers do value high quality products from non-targeted sellers and that this valuation actually is increased after the implementation of this monetary incentive and finally the last piece of the puzzle is that we run the main DID specification separately for the high and low quality products by non-targeted sellers and we find a strong evidence that non-targeted sellers reallocated their supply from low quality to high quality given the rising demand for high quality products from consumers so let me conclude in this paper we study the effects of a targeted incentive we find that besides the targeted sellers non-target sellers are also more likely to adopt the promoted behavior and increase their quality provision and the reason for that is that non-target sellers experience a larger demand expansion which motivates them to enhance their quality and therefore the overall quality provision on the platform can be increased as an equilibrium result so this is all I have thank you very much thank you so much my new um I didn't get any questions so far um so so so in the absence of questions I I imagine people are leaving them to the end so so we have to we might move to to the next talk which is by Rob with some some OG some good advice from Johannes Pohnen Robert yes so I'm sharing my slide let me thank the organizers for this opportunity and also you'll see for all the comments and questions we have already exchanged so I will talk about deceptive products on platforms joined work with Johannes Jona so many products sold on platforms include additional fees um examples are flight comparison websites like sky scanner google flights to your favorite and luggage fees price comparison websites like ebay and amazon shipping fees um even ticket platforms like stop hub and ticket master and service and um any of these fees um are obfuscated not very clearly presented on platforms and all around the world uh policymakers have been increasingly active regulate investigating and sometimes regulating um these uh hidden additional fees um like one example I like is the EU has recently pressured rbmb into uh displaying cleaning fees up front and cleaning fees are fees that we will call salary fees that goes to the post not even to rbmb and still and rbmb seem to try to shroud it by showing it only later in the booking process it's also called reprise so a first question is why the platforms obfuscate fees that they do not even earn themselves and uh more generally uh we have I think a good reason to to believe that when it comes to additional fees uh large online platforms are private rule makers because they can decide whether to shroud or unshroud either reveal these additional fees so they would be in a good position to induce transparency if they wanted to sometimes they don't want to and our more general research question is what are platforms incentives to create a transparent marketplace so um I will only have time to flash the key ingredients uh of our model the first one is that by now we have solid empirical evidence that on many markets at least part of the consumers um seem to um ignore or forget about additional fees whenever uh whenever they are hidden and so we will um build on government lives on seminal paper uh when modeling naivete um so sellers choose a base price which is observable to everyone and an additional fee which is not so part of the consumers will be naivete um who just forget about the additional fees whenever they are hidden they falsely believe them to be zero and the other consumers are sophisticated who can anticipate these fees and avoid them by a costly effort the second main ingredient of the model is that the sellers compete on two-sided platforms um um and uh so we have a monopoly platform in the baseline model um we use a simplified version of armstrong uh seminal model and um so importantly the the positive cross group externalities are endogenous in our model they come from the interaction of sellers and buyers uh and they are affected by shrouding and third um I think we have good reasons to believe that it is uh the platform's design choice whether they want to hide or reveal the additional fees and uh we will distinguish two types of additional fees seller fees and and and platform fees seller fees are like the cleaning fee I mentioned that the seller set and the seller's burn various platform fees are fees that the platform set and earn but in most cases uh we believe it is a realistic assumption that it is the platforms we choose to shroud around shroud them and um so um it is the interaction of naivete and two-sidedness that creates a novel mechanism and which in turn creates novel insights in our in our paper um so the main finding is uh the platforms have quite strong incentives to to hide additional fees and I will very quickly show you two results that support this statement uh the first result comes from um comparing um our model where the platforms decide about shrouding and unshrouding um with a benchmark model where uh platforms let the sellers decide they're to shroud around shroud this is similar to an unintermediated uh interaction between sellers and buyers and what we find is that it is actually the platforms who have stronger incentives to shroud seller fees than sellers themselves which is not surprising at first sight because it is the sellers who earn the additional fees um and very quickly the intuition is that if you start from a situation where um unless if our platform all sellers shroud if one seller uh unilaterally deviates and unshrouds reveals the additional fees then it will attract buyers by being transparent but if we reduce overall demand because some people will realize that the product is more expensive than what they thought and because of the cross group network effects the platforms will actually suffer more from a reduction in overall demand than the sellers so what happens is basically platforms uh coordinate shrouding because they want to prevent sellers from competing in uh transparency because they want to avoid a loss loss of demand so from a regulatory perspective these results suggest that uh the existence of platforms may may verse a known transparency in addition um we believe that uh our results nicely connect to empirical evidence uh there is a recent working paper uh that uh describes uh field or well that's a field experiment uh on stock which is an even ticket platform and we chose that shrouding a 15 percent additional fee which was a service fee actually boosted revenues by 21 percent so I think the order of magnitude is quite quite remarkable we have indirect evidence that online platforms shroud a lot because they used repricing a lot which is a form of shrouding and there is one piece of direct evidence from a 2012 report for the EU commission that found that two years after the introduction of an all-inclusive of the all-inclusive up from pricing rule for airlines uh almost the share of online platforms who violated this rule was almost twice as high as the share of airlines who violated this so in this case platforms really seem to shroud more than sellers and um our second main result relates to platform fees so these are the fees like service fees that the platform sets and collects and we find that uh the fears are the competition among sellers on the platform the stronger incentive this platform has to shroud its own fees um so we see this as as as a novel adverse effect so platforms have strong incentives to hide their own fees exactly because they induce a competitive marketplace which is also in contrast with the common argument that online marketplace is always facilitate product comparison um so from a regulatory perspective um i think i think frank will talk about um a model and a possible intervention that intensifies competition among sellers on a platform so if you are to um if regulators are to introduce such such an intervention our model suggests that they should not forget about additional fees either because that would backfire so to conclude in this paper we explore platforms incentives to design a transparent marketplace we have two results that both indicate that they have quite low incentives to do that we believe that this nicely connects to evidence on repricing and um finally i would just like to mention that i think we have quite a few applications both avoidable and unavoidable fees repricing and although in the baseline i told you about it's a monopoly model we have one extension about we have many extensions one of them is about competition between platforms that might might not help transparency either so this is it thanks thank you so much robert for this great talk uh so we do have a few questions on on the chat so alexandro the coroner is asking uh there's a result that's platform have an incentive have an incentive to shroud depend on whether they charge sellers and buyers um yes so so in our model um they are allowed to charge uh both sellers and buyers like in the answer model you know the membership fees um and um um in some of in some of the parameter regions the membership fees for buyers endogenously becomes zero which is quite realistic in online markets so buyers don't have to pay to use the platforms um but they could they could charge them yeah no my question what thanks my question was about uh and i think this clarifies it the intuition would be that if you can extract surplus from buyers then you might want to actually improve transparency maximize competition but you're saying that that's not the case that that's your effect won't be killed if you can charge food yeah okay this this is not what um this is not there is this effect but there is this additional effect coming from the two-sidedness of the market right you want to extract actually so you want to shroud around shroud based on whichever leads to a higher buyer surplus because that even if the price is zero for buyers by the two-sidedness um you can extract more money from the seller side so this is exactly the core core mechanism we have but you don't have time to talk about that thanks thanks for the question and there's a second short question bushing is is wondering that the platforms they have no application of concern yes i see yes i um this is this is a this is a good point we have we have talked about it actually so it's true that we we don't talk about reputation at all we abstract away from that but um when it comes to um additional fees it's not really clear if uh you would blame uh the platform or you will blame a seller so imagine that you're in a hotel booking platform like booking.com and uh you arrive to the hotel you have to pay a resource fee so will you blame booking.com for not displaying the resource fee or will you blame the hotel the seller um because they charge the resource fee so for this reason we think that reputation concerns are of second order import but this is this is a good point okay so thank you so much Robert for the great talk and for the discussion uh so we are right on time for the second to last presentation by Ron Berman in joint work with Yuval Helen Ron thank you um hi everyone can you hear me okay i can excellent um and you can see my slides great so um thanks for inviting me to present this talk today my name is Ron Berman at the Wharton school uh this is joint work with Yuval Heller and today i'm going to speak about why firms often prefer not to correctly measure the response of their payoffs to all sort of actions that they take and we call this uh solution concept or this result the naive analytics equilibrium so we started uh basically with motivation which is kind of empirical observation that a lot of firms use observational data to to do analysis and they choose very very simple analytics they just run maybe a linear regression they just do simple averages they often do not correct for the generality of the variables they are reluctant to run experiments they say it's too hard um and there's another set of observations and this is comes from literature but also from popular press that says firms often overestimate their advertising effectiveness so they spend a lot of money on advertising but actually it seems to be not very effective definitely online and in some cases when friends don't correct uh for the generality of the variables they also underestimate price elasticity this causes them to maybe incorrectly price their products so we have three research questions uh first of all is why do firms maybe kind of choose or converge to playing kind of this using this naive types of analytics we are trying to explain the direction of the naivete like do we actually can explain why firms overestimate advertising effectiveness and maybe underestimate price elasticities and maybe other phenomena that we observe and uh is it beneficial for the firms and finally we're going to ask what happens in equilibrium the firms uh overadvertise underadvertise the too low prices too high prices and what happens to their payoffs so this is a theoretical paper uh we have um actually the model itself it is not too complex i'm going to explain it fully there are n players each players has a payoff function pi times on two things there's the action of the firm you can think about other setting the price or maybe the advertising budget or something like that and the demand and the demand that the firm observes or realizes depends on the actions of all firms in the market um and we assume that the demand is not fully known by the firm and what do we mean by not fully know we assume there's a two-stage game um in the first stage the firm um doesn't really know exactly uh how its actions xi affects the payoff so they hire an analyst the analyst does something with data and the goal of the analyst is to estimate the response of demand to the firm's action and then the firm takes the estimate of the analyst assumes it's correct sets the optimal let's say price or advertising based on that now what would firms want to do firms would optimally want to solve the the first order condition as if they knew the correct profit function and this would depend on three things there's going to be the direct effect of the actions of the payoff there's going to be the indirect effect through uh affecting the demand and what we're assuming is we're assuming that firms know if I increase the price what happens to the margin of my profit um or if you know I sell this product or more product what happens to my payoffs or my profits but what we assume the firm doesn't know is a derivative they don't know exactly how mixing advertising affects uh demand or maybe changing prices affects demands because it also depends what other firms do it depends on uh unobservables in the market etc instead they hire this analyst and the analyst can be biased uh the analyst instead of estimating dqi to dx i which is what the friends would like the analyst to do they might have a bias um and they estimate alpha i time dqi to dx i and the firm takes that as if it is true and solves what we call the bias first order condition and in the bias first order condition the firms choose the action that maximizes their profit as if the estimate from the analyst is correct now you might be asking yourself how can analysts make these mistakes so actually it's pretty simple so for example suppose an analyst wants to estimate price elasticity they tell the employees you know what just change the prices a little bit on on a few days give a discount and i'll use that in my regressions well if the employees are let's call it lazy or they choose to give discounts on day with lower demand you will get this correlation and this will create indigeneity and this will have bias in the analysis a slightly more sophisticated example if two competing firms somehow set prices the same way let's say they give discounts on weekends you will also have the same issue um that you will have a bias analysis but the important part of this model is that the analyst never knows and the firm never knows that what they estimate is inconsistent with the data their estimate is always consistent with the data given the data that they have it's not that they have a misperception of their payoffs they're actually just incorrectly estimating what's going on because it fits their data so solution concept is two parts um there's the equilibrium when the firms set their actual let's say they're set their prices we call this an alpha equilibrium we require two parts out of it the firms need to be setting the let's call it the prices or the actions xi to solve their bias first order condition and also the bias second order condition needs to hold but the crux kind of of this equilibrium comes from what happens in the first stage and the second stage together an analytics equilibrium is a set of biases by all firms and actions by all firms where in the second stage they pay an alpha equilibrium but in the first stage what happens is the asset they chose maximizes their profit if they deviate to a different level of bias they will deviate to a different equilibrium in the second stage that will lower their profits let me just an example suppose firms play standard differentiated downtrend duopoly so here the demand is going to be the intercept minus the price times the price sensitivity of consumers to my own price and also if the other firm increases the price my demand increases the profit is going to be very simple my price times my demand we prove in the paper that in this case is going to be a unique naive analytics equilibrium all firms in this case two firms are going to underestimate their demand elasticity this is a game of strategic compliments this causes them one firm increases the price the other one also increases the price because we're trying to competition as a result the profit is also going to increase and the payoffs are going to retro dominate the ones in a national equilibrium so then we wanted to ask ourselves can we generalize this result and can we say something about general games with more players and is there like a pattern that we can find into when would firms under overestimated demand sensitivity and what would be the results on profits etc so because it's a short presentation i'm just gonna on one example take a look at advertising here we have a game of advertising with positive externalities in a game of advertising with negative externalities the positive externalities the advertising strategic compliments with negative strategic substitutes what we show is that depending on the external one firm on the other and depending on how my actions affect my own payoffs when i increase my advertising it increases my payoffs what we will observe is that both firms will overestimate their demand sensitivity in respect to advertising which is what we actually observe in reality this will cause them to over advertise with compliments they will make higher payoffs with substitutes they will make lower payoffs it's if the equilibrium is symmetric yossi emailed us before this conference it was kind to show us actually that he has two papers that has actually very very similar results the main difference is that in those papers the firms misperceive their payoff functions while in ours they're trying to learn from the data and because they are naive in their analytics they incorrectly learn how their demand responds to actions so just to conclude what her model allows us to do is explain three seemingly unrelated phenomena i talked about overestimating price elasticities sorry underestimating overestimating advertising effectiveness we're also showing the paper that this can explain overconfidence when you have teams of people trying to create a joint project this is very similar in outcomes to the results from the delegation literature but the mechanism is different it's not that the firms try to serve different payoff functions for employees to do something here they're coordinated they're trying to learn the truth they just incorrectly do that the insights from this analysis actually have some empirical implications when we do counterfactual analysis in structural work we assume firms know the demand elasticities and they set prices optimally but here if they actually converge to almost on purpose misestimating the elasticity they will set different prices and we need to think what it means when we do counterfactual analysis and finally we also now have kind of this trend of telling firms run more experiments be more accurate measure everything more correctly but actually this paper shows that in this case their profits are going to suffer and actually it's not necessarily better to invest in better analytics maybe you want to remain naive thank you for listening to the talk there's more in the paper there's more in the short video i presented for the for the conference and i would love to answer any questions if you have any Ron thank you so much for the great talk i'd like to postpone questions to the end Yossi has a comment on the chat but i'd like to postpone the questions to the end because because we are we only have five minutes and there is a there's a last paper to be presented by Frank Schluter so so frank the floor is yours thank you so much thank thank you very much much do you see my um can you see my slides i do i do and i hear you great great um so perfect thank you very much for the invitation to this very nice conference and also thank you very much to Yossi for for the introduction um my name is Frank Schluter i'm a PhD candidate at the Dysseldorf Institute for Competition Economics and i'm happy to talk about about my paper managing seller conduct in online marketplaces and platform most favorite nation courses in this paper i show that these platform most favorite nation clauses can reduce the incentive of a digital platform to ensure strong competition between the online seller set are active on its marketplace there's this huge and ongoing concern how digital platforms such as the amazon marketplace or booking.com organize their marketplaces and whether we need stricter regulation for the digital economy um just to give you one example regarding these concerns european commission amalgamated astya notes that there are a few gatekeeper platforms that act as private rule makers for the marketplaces that they have created so platform behavior or the rules at the platform set have an important influence on the functioning of digital markets and whether such a platform wants to ensure strong competition between the seller set are active on its marketplace is arguably one of the main prerequisites for consumers to benefit from low and competitive prices when purchasing online at the same time we have encountered cartel cases involving online sellers on exactly these digital platforms which suggests that this competitive marketplaces that we might expect are not always provided in these environments and i want to contribute to this broader debate with a specific focus on platform most favorite nation clauses and answer the research question how do these clauses affect the incentive and ability of a platform to ensure competition between online sellers let's consider the example of the hotel market you can see a hotel that wants to sell rooms to consumers and it can do so via different distribution channels can use a platform like booking dot com but it can also sell directly to consumers for instance via its own website or different distribution channels that you can see here and in principle the hotel is free to charge different prices on all of these distribution channels in order to maximize its profits but in this environment platforms like booking dot com have imposed these platform most favorite nation clauses that are at the core of my paper and they are a contractual restriction not to offer better prices or conditions on another distribution channel than on the platform itself so if booking dot com imposes such a clause and the hotel charges a price of 100 on booking dot com it is not allowed to charge a lower price on its own website and these clauses have attracted substantial antitrust scrutiny and current discussions in europe regarding the digital markets act also propose prohibiting these clauses altogether now i study this environment in a stylized theoretical model of digital platform markets where there are two sellers and they can reach consumers via two different distribution channels they can reach consumers via platform but they can also sell directly to consumers for instance via their own website and the platform uses a so-called agency model which implies that it receives a commission payment for every intermediate transaction on the platform and it's the online seller sets that's a final retail process um it's easiest to explain the results for the case of per unit commission rates which is why i focus on this case here in the presentation but in the paper i also analyze revenue sharing commission rates and if you're interested um you can have a look at the paper now it's briefly briefly consider how a platform most favorite nation clause affects um online seller pricing in this situation absent these clauses online sellers can react to the commission payments on the platform by charging a lower price on the direct channel and this allows them to divert sales to the direct channel where they do not face commission payments and find it more profitable to serve consumers now this pricing structure is forbidden by a platform most favorite nation clause and induces online sellers in this model to charge uniform prices across distribution channels i want to use this model to briefly explain the two main results of the paper the the first main result is the platform's incentive to ensure competition between the sellers and how it's affected by a platform most favorite nation clause particularly i show that a platform may prefer a form of non-competitive or monopolistic environment between online sellers if it is allowed to impose a platform most favorite nation clause and for this part i suppose that online sellers can coordinate on the joint profit maximizing behavior the monopolistic case and compared to the non-cooperative behavior in this situation the competitive case if you want so and i show that absent the platform most favorite nation clause the platform unambiguously prefers sellers to compete and the main reason for this is that intuitively for a given commission rate the platform wants to have the maximum transaction volume on its marketplace at least for the case of per unit commission rates um which is the case that i focus on um and this is exactly what is achieved by seller competition for the platform it reduces the prices that online sellers charge and this increases the transaction volume on the platform and as i said this is the main reason why seller competition is profitable for the platform now perhaps surprisingly this is the result that changes quite fundamentally with the introduction of the platform most favorite nation clause and here i show that the platform may prefer monopolistic seller behavior if it is allowed to to impose such a and the main reason for this result is that um with with these clauses monopolistic sellers actually accept higher commission rates in equilibrium than competing sellers do and this increase in the commission rate can render this monopolistic seller behavior the more profitable conduct for the platforms and seller competition despite the negative effect of online seller collusion on the transaction volume that occurs by yeah so let me also briefly talk about the second result that i want to explain the platform's ability to stabilize collusion between online sellers and here i show that a platform most favorite nation clause gives the platform the ability to stabilize collusion between online sellers and that it also finds it profitable to do so and this approach is mainly motivated by the cater cases involving sellers on digital platforms that i've mentioned before and it is a quite direct mechanism that allows online sellers to sustain this coordination on the joint profit maximizing behavior that i've assumed for the first result and here the main reason why the platform can stabilize collusion is that it can charge a comparably high commission rate for which colluding sellers are willing to list on the platform but the participation constraint of competing sellers would be violated and they would decide to de-list from the platform and compete rather aggressively on the direct channel of the law so in that sense competitive profits are rather low in such a situation and this stabilizes collusion in the sense that it leads to a decrease in the critical discount factor which is necessary to sustain collusion as a sub game perfect equilibrium now let me just briefly note that there is empirical research on these platform most favorite nation clauses as well and that suggests that platform most favorite nation clauses indeed affect the listing decisions of online sellers so this appears to be a relevant choice dimensions for online sellers if these clauses are used now let me briefly summarize and i argue that based on this result that my paper establish a novel theory of harm regarding platform most favorite nation clauses linking these clauses to reduced competition on the level of the sellers and thereby adds to existing concerns regarding these clauses that typically focuses on reduced competition on the platform level so thank you very much i'm happy to answer your question okay so so thank you so much for the great talk frank um so Doshin has a has a comment he says then according to the revealed preference argument sellers should prefer the platform mfm is it the case um so if for the case of colluding sellers it is so if if the conduct of online sellers is collusion or in this monopolistic setting platform most favorite nation clauses is clearly profitable for the platform but if there is competition between online sellers this participation constraint which is binding in this case might be very tight for the platform and this might then be even detrimental for the platform and this is also part of a of a recent paper by Johansson Veggie and they also also spread this out did you analyze the case of competing platforms and how the how the multi-home your single home of consumers affects result um so the industry structure that i consider as one platform and one direct channel and i don't have a second competing platform or i don't i haven't analyzed this yeah okay so so i'd like to thank you frank and and also thank yossi for his great introduction and uh and all the speakers for this great session we are five minutes late so i think that's that's a good performance and uh okay so so thank you very much and have a good evening