 Topic of this paper is competition on vertically integrated platforms and e-commerce platforms, and you could think about this as especially Amazon, but could potentially apply it in other settings that I'll discuss in a bit, serve at least two distinct functions. First, the sort of maybe the most well-known function is that they host marketplaces which match consumers with external sellers of various goods. But what they're also increasingly doing is producing their own house brand products which they sell directly to consumers on the marketplace in competition with external sellers. And now this dual role leads to a variety of potential conflicts of interest, in general the idea being that the marketplace oversight that the platform wields gives them potential the ability to advantage their own products over external sellers. Now there are a number of different directions in which you could think about conflict of interest here and the one that we're going to focus on for this paper is going to be the use of proprietary demand data. So what we mean by that is that by overseeing the platform, the sorry by overseeing the marketplace, the platform then gets to observe which third party products are selling well, how often customers reorder these products, so some notion of the kind of stability or longevity of demand, as well as what products consumers or folks are searching for. And all of this data can both hypothetically and in practice be used to target the introduction of house brand products in particular high demand segments. And so the kind of potential worry here from a competition standpoint is this practice which we'll refer to as insider imitation for this talk may potentially harm innovation or the introduction of new products onto these marketplaces by sapping profits from so-called long tail firms that sell mostly or only on a single platform. Now this is not just a hypothetical practice given the audience I probably don't need to motivate this much further but you know just to hammer this home. Last year the EU brought a sort of preliminary notice or published a preliminary notice that they were planning on bringing a suit against Amazon for breach of interest rules. There were actually a couple distinct practices that they noted in this notice but one in particular was specifically the use of this non-public business data in order to target competing products. Now you know just to kind of motivate this exercise even further you can ask kind of what's new you know why is this kind of private level competition particularly concerning for online platforms when this is a really commonplace practice in brick and mortar retailers like you know grocery stores department stores and so on. You know there's evidence going back at least to the 1870s with Macy's that department stores have been private labeling for a very long time and it's also very clear that the sales data in these brick and mortar stores is also routinely available. Now one answer to the what's new is that maybe this has always been a problem in all venues but some of the horses left the barn for brick and mortar but people are interested in regulating online so we have the chance to do something about it. I think that's a perfectly good answer but we think that actually another distinction that might make this a particular even theoretical interest is this idea of unlimited shelf space in e-commerce. So large retailers essentially by definition can carry only established products with a proven track record because they need to carry these products at scale. By contrast online marketplaces like Amazon can carry any product and in fact often pride themselves on having fewer virtually no barriers to entry and Amazon has even had problems with this where the kind of selling of stolen products has been difficult specifically because Amazon doesn't want to raise many barriers to sellers listing on the platform. And as a result of this of these this kind of low barriers to entry survey data suggests that most Amazon sellers are quite small often only make a few thousand dollars a year in revenue and sell either mostly or exclusively on Amazon. And so in terms of kind of what's new and motivating this exercise we think about online marketplaces as providing a novel venue for introducing and experimenting with new products and that raises distinctive antitrust concerns. So the basic question that we're going to tackle in this paper is how should a regulator think about restricting the use of marketplace data in order to kind of improve innovation and also to kind of maximize social welfare. And the key assumption that we're going to carry with us throughout this talk is that the platform is itself a strategic actor with commitment power. So we're going to assume that the platform can choose what we'll call an imitation policy strategically and can kind of commit to it in advance of the entry decisions that are made by entrepreneurs. And so the platform also makes a trade off that's sort of at least qualitatively similar to the regulators where it's going to trade off the benefits to its marketplace business of more intra-bound entrepreneurs against the benefits to its product division of imitating successful products. And so in order for regulation to serve a role it's going to have to do something beyond just solving a holdup problem of allowing the platform to commit when it couldn't otherwise. We're going to kind of take as given that the platform can solve that problem for itself by establishing a reputation. And we're going to then ask okay you know how does that leave incentives misaligned even with commitment power and can a regulator do something about that? Okay so this is a very short talk so I just want to kind of you know highlight that there's other work that's been done on closely related issues of sort of a vertically integrated online platforms and data usage even quite recently. And in fact I think most of the papers I'm citing have at least one author who's in the room right now. So there's been a lot of interest in recent work about regulation of so-called dual-moting or self-preferencing. Dual-moting is just the idea that a marketplace should can sell its own products at all and so papers have studied what would happen if you outright banned the sale of product of private little products on these online platforms. And also they have self-preferencing that you can somehow buy a search results or or recommended results to privilege private label products. There's also been a little bit of work about regulating the use of data collected from consumers which is sort of a complementary issue to ours. And so our contribution just to keep it very brief is we think about a slightly different issue which is not about kind of a restriction on vertical integration per se but about a regulation of collection and usage of merchant data in particular. So that's what we think of as being the kind of particular niche in which our paper sits. Okay so let's move right on to the model. So the models are quite simple. There's going to be a small number of players. There's going to be a platform and that platform is going to do a couple things one of which is to host a marketplace for an entrepreneur's product and then there's going to be an entrepreneur and that entrepreneur is going to develop a new product and potentially sell it on the marketplace. The platform is then going to have one additional thing that it can do which is that it can observe a product that's been introduced and it can imitate that product. Then there's going to be a regulator and that regulator is going to implement data protection rules and I'll tell you more about those rules in a bit. So there's going to be a couple key product market variables that are going to affect the entrepreneur's decision to enter and the platform's decision to innovate or to imitate. The first is going to be an innovation cost which is going to we'll denote by little k and it's going to be drawn from some distribution whose features aren't that important and the idea is that this innovation cost is going to determine whether a particular entrepreneur enters and it's going to be the private knowledge of the entrepreneur. The other important product market variable is a demand state which will denote alpha that's going to be drawn from some other distribution whose features will be important and we'll talk about that later on and this is going to basically determine how big the market ends up being for newly introduced product. Now the entrepreneur's problem is they look at their innovation cost k and then they decide whether they want to pay that cost in order to innovate by developing a new product and then if they decide to develop a new product then they're going to get some profits from selling that product in the marketplace. We're going to denote these profits by pi e for entrepreneur. They're going to depend on both the demand state and they're also going to depend on how much competition the entrepreneur faces in the marketplace. So we're going to assume that the demand state is sort of a positive indicator of profits so the higher demand is the higher profits are but we're also going to assume that the entrepreneur doesn't like being competed with very naturally so we're going to call the state where there's no imitation by a private label the monopoly state and when there is imitation the duopoly state and the entrepreneur doesn't like competition. Can I enter please Eric sorry we have a couple of questions so about the first one was about the commitment power of the platform so Doshin is asking Doshin and asking what is the rationale for the assuming that the platform can commit to its imitation policy. Okay so were there a couple of questions did you want to say all of them? Yeah just that one I think the other one we can keep for the maybe discussion because it's about a little bit more like how to think about it's about the you know I can ask it now so that you can think about it along the presentation and decide when to answer. So this is from Jack Kramer he asks amount of products that Amazon sells under private labels how many are really innovative trash cans are not so. Yeah okay good so to answer the first question about commitment I think this actually might be something that could also be good to to defer a little bit to the discussion section we think of there as being two reasons to focus on on commitment one is we think about this is just sort of a benchmark problem that says okay well there is always the potential for regulation to solve a holdup problem we kind of understand how that could work in a variety of settings but if we want to understand kind of the real conflict of interest between the platform and the regulator then we think about commitment as sort of giving us the the benchmark we need to really understand what the conflict of interest is there we also think it's reasonable that a large platform like Amazon sets policies that it you know it keeps consistent over time that seller that you know potential sellers in the platform internalize and that it can really develop a reputation by its practices and you know in fact it has developed in some sense a bad reputation for some of the the putative practices like you know downgrading sellers who don't use their fulfillment services for listing the buy box and things like that so you know maybe we can discuss more in the discussion section you know whether we think there might be a problem of it's kind of secret deviations from this but again we think about this is a good benchmark to understand conflict of interest then to answer jock's question um sorry can you repeat jock's question so it you can also answer it later on i mean the amazon sells under private labels many products but how many are they aren't really innovative yeah well okay so actually i think to the extent we think that amazon the private level products are not innovative we think about that is actually that that sort of fits with our model but the idea is specifically that the platform in our model is not an innovator all they can do is look at something that's already listed on the platform like say a trash can and then decide whether to to introduce that for itself now presumably for a trash can you know amazon doesn't need to see uh an entrepreneur enter in order to know that trash can exist um but you know as we see that amazon there are lots of different variations of products that you could think about as being innovative in the particular features or you could think about whether they're sort of a particular marketplace for for you know to sell trash cans online for instance as opposed to in a physical store is also something that there's some innovation in being willing to kind of take the risk of of experimenting with that um now maybe that's something that you again could think about amazon is being able to do without seeing an entrepreneur but we do like to think about there being a lot of products where amazon couldn't have uh wouldn't have had kind of a competitive advantage in thinking about this particular idea um and they really need to see an entrepreneur do it first and then they can imitate it at low cost if they want to um so one example that we like from a wall street journal investigation about these practices um was this small kind of two-guy company that introduced a trunk organizer um which is just basically a set of partitions that you put in the trunk of your car that allows you to kind of organize whatever things you put into the trunk um and that sold very well for a few years in amazon and then amazon came in after a couple years and imitated it so we think about that as the kind of thing that it's an innovation of the sense that it's not a product that that was kind of immediately obvious that you would kind of want to develop um and once the entrepreneurs came on and did it and it was a success then amazon kind of woke up and noticed it uh okay good so i don't want to keep you you know so slow at the start but uh i think that question is important to ask here uh about the exogenous cost of innovation how to understand this uh innovation costs being exogenous coming from lewing one uh i mean i guess the easiest and most literal way to understand it is just that there are going to be some products that are relatively costly to develop and some products that are going to be less costly to develop and so you know we don't model the specifics of individual products here but you could think about this as capturing some important amount of heterogeneity in the types of products that are being introduced so that's the way that we'd like to think about it here but you could also literally just think about it as they're also being heterogeneity in the efficiency of different entrepreneurs and this is really the simplest way to capture unobserved heterogeneity that will produce some smooth response and how much innovation happens in response to a policy change by the platform okay go ahead yeah all right great okay good so now i've described the entrepreneurs problem um which is quite simple it's really just a static decision about whether to innovate or not uh there we go so now next up we have the platform's problem um so if the entrepreneur decides not to innovate then the game ends and the platform collects zero from this product that didn't enter um if the uh entrepreneur does decide to enter then the platform gets to collect a stream of flow profits um which we like to interpret as commissions on the entrepreneur's marketplace sales um also at any time the platform may choose to introduce a competing product if it introduces this competing product it pays an invitation cost kp um which we're not going to model as random because it's from the point of view of the platform it's not uh and then its own flow profits which are going to be kind of an amalgam of the commissions and own sales are going to increase um from the monopoly level to some higher duopoly level and we're keeping all of the details of this competition very reduced form you know you can choose your favorite model of product market competition to get these um we're going to place very few restrictions on exactly what these numbers are um so just uh to kind of summarize the basic timeline the entrepreneur is going to draw their innovation cost and decide whether to innovate they don't innovate the game ends if they do innovate the product is introduced the demand state for the product is going to be realized um and at least in the benchmark models can be observed by the platform uh the entrepreneur is then going to sell their product over an infinite time horizon the time horizon here is actually not going to play much of a role in the benchmark model but will be important under regulation um there's going to be some discount rate which will not play much of a role and then the platform is going to choose a time at which they want to imitate uh potentially as a function of the demand state um and so you know whenever it chooses to imitate it pays this fixed cost and then the market structure is going to switch from monopoly to duopoly uh now as I mentioned before we're going to give the platform commitment power it's going to commit to an imitation policy and what we mean by an imitation policy is very simple it's just a mapping from the demand state for the product to a time at which the platform wants to imitate the entrepreneur is then going to get to observe the platform's imitation policy before it decides whether it wants to to innovate as I mentioned this commitment power is going to allow the platform to internalize the impact of imitation on innovation and so really going to allow us to clarify what the conflict of interest is between the platform and the regulator now the first result that I'm going to show you guys is going to be a very kind of blunt sort of a regulatory instrument the regulator can impose what we'll refer to as an insider imitation ban um and the insider imitation ban is basically going to sever this link between the demand state and the the imitation time so we're going to model the demand state very simply as being perfectly inferrable for marketplace sales data by default absent any regulation the platform gets to pass that data from the marketplace to its product divisions um and so what we're imagining in our our sort of first pass at this problem is that the regulator can mandate a Chinese wall between these divisions so the divisions just can't share data with each other this is something that's very common in in financial regulation uh and so the regulator's problem formally stated uh is should the platform be allowed to condition its imitation time t on the demand state alpha um so for the purposes of this talk what we're going to look at is uh whether this policy would increase the amount of innovation um and in the paper we do a sort of more nuanced analysis where we incorporate gains from competition um into this um but at least if the the the regulator's interest in new products is large enough um then the insights of from maximizing innovation are going to continue to hold um okay so also I'm I don't want to belabor this too much just to keep things simple um all of these profit variables that I've written earlier generically as functions uh we're actually going to take to be linear um when they're linear we can without loss normalize the demand so the average demand is one um and this linearity isn't crucial for much of our analysis but it really allows us to take our results more cleanly and powerfully it allows us to talk about uncertainty about profits independently of average profits uh given that potentially all these profit functions and non-linear functions you can't really renormalize things so that all the average profits stay the same as you move the demand around so linearity lets us do that okay so let's just talk very briefly about what the how the platform is thinking about designing an imitation policy absent any regulation and then I'll come and tell you how this thinking is going to change under the regulation so the platform's problem has two kind of simple components um one is it wants to kind of do a uh a gross scale trade-off between innovation and imitation which you can think about as targeting a particular innovation threshold kappa this is going to be the maximum fixed cost that an entrepreneur would be willing to pay to enter the market um but then it's also going to be able to tailor his invitation uh to the particular demand state and so it can design uh an imitation policy that optimally implements you know a given target threshold um so the second part of the problem you can think about as a classic sort of cost minimization step so how do you optimally imitate kappa in a profit maximizing way um and it turns out that absent regulation you always use a sort of bang bang policy um and the bang bang policy says in good demand states you annotate immediately in bad demand states you never imitate um so we could have written the no regulation version of this model without as sort of a static problem without any time dimension and just thought about it as which markets does the platform enter or not and you wouldn't really lose anything by doing that um and so I just want to introduce one kind of really crucial concept uh that both motivates this result but also will be useful for explaining our main results so the reason why the platform invitates only when the demand is high is that it thinks about this conversion rate between the entrepreneur's profits and its own profits so the way it siphons off profits from the entrepreneur is by imitating um and it turns out that the rate at which it convert it can convert the entrepreneur's profits into its own profits is going to be increasing in the demand state um basically because it it the fixed cost becomes a lower portion um of the the variable profits and so it gets to siphon off the entrepreneur's profits most efficiently in good demand states and that's why it's going to concentrate all of its invitation um in good demand states you know holding fixed a given amount of profit that is siphoning off and I'll skip over this last point um so then once the platform uh you know has chosen the demand uh threshold to implement a given kappa then it has to optimize over this kappa I'll call the optimal kappa threshold kappa star um and it's going to trade off two very natural forces on the one hand if you increase kappa then uh or sorry if you uh if you decrease kappa by imitating more then your profits are going to increase by imitating additional products but you're going to suppress entry and so you're going to have uh less entry of high cost entrepreneurs and and that you know the market is going to get established less often um and it's going to turn out that that these two uh margins the kind of uh you know we think about the first as like being a a marginal effect where you get a little bit more profit from all the existing products and the or that's that's sorry that's a marginal effect the inframarginal effect is that you get less entry um and there's going to be a distinction between the conversion rates and in each of these two uh parts of the first order condition um so if we think about the additional imitation of existing entrants then that's going to generate profits at what we'll call the marginal conversion rate and that's exactly going to be how you convert profits if you take a little bit more demand away from them or you or you imitate them in a a few more demand states um but if you get new entry of entrepreneurs then that's going to generate profits um at a different conversion rate what we'll call the inframarginal conversion rate and there you get to imitate them not just the marginal demand state but at a bunch of demand states above the margin and so you actually get to convert profits more efficiently for all of these kind of new entrants um versus for for an existing entrance and this will play you know this both determines exactly what the kappa star is but I won't tell you the details of that but it's going to play a really key role when we understand what regulation does okay so now let's just talk about how regulation impacts things so we're going to put in place an insider imitation ban this isn't going to prevent the the platform from imitating a product but it's not going to allow it to condition on this proprietary demand data um you know you can under the ban the platform's problem just reduces to a no uncertainty benchmark in this benchmark the demand state is going to just be fixed at one and because everything is linear so the expectations just take all the office to one um and the conversion rate is going to end up being fixed so you're just going to get this convert this fixed conversion rate um lambda bar which is going to be equal to the the conversion rate at the demand state so the question is how does this change in conversion rate affect the profitability of imitation it's going to turn out it's going to do different things to this marginal effect and this inframarginal effect so the marginal effect is that the conversion rate for new entrance is going to move from this average conversion rate uh to a different average conversion rate and I'll tell you on the next slide about what the comparison between those things is and then there's going to be an inframarginal effect which is the conversion rate for the existing entrepreneurs who are already entering is going to move from this kind of marginal conversion rate to this average conversion rate um and so it turns out those effects are going to be very different um on the margin and on the inframargin so the direction of the marginal effect is is going to be clear um this kind of a the conditional average demand state these are the average of all the demand states in which you imitate is always going to be higher than average because you're only uh you're only imitating in the good demand states and so the conversion rate for new entrance drops another regulation um and this force is going to mean that the platform has less of an interest in admitting new entrants because you don't get to convert their profits as efficiently because you can't target your imitation um and so that force is actually going to tend to push the optimal kappa down but the inframarginal effect is going to be less clear um so the marginal demand state need not be comparable to one it could be greater than or less than one um and so the direction of the effect is really going to depend on whether unregulated marginal demand is going to be above or below the average demand state um and so the the kind of net of these effects is that if the marginal demand state is low so you're imitating a lot of products including ones that are below average um then these two effects are going to point in the same direction um and this is going to be kind of the wrong direction from the regulator's point of view that this is going to be a bad outcome for the regulator where the platform is actually going to choose to imitate even more products when it can no longer tailor to demand on the other hand if the marginal demand state is sufficiently large then the inframarginal effect is going to oppose the marginal effect and can overwhelm it and that's the case in which the regulation is actually good and it increases innovation um and so our main results very briefly are just going to identify sufficient conditions on the distribution of demand which are going to identify the net effect of the regulation on innovation um and these are basically going to to affect how this marginal demand state alpha star is going to move around um so the concept that we use to kind of quantify this is what we call the upside demand potential um the upside demand potential is basically the kind of amount of demand that's in some given right tail of the product uh or of the demand distribution so we're going to call this uh we're going to refer to this relative to this function we're going to call it capital phi uh we're going to call this the alpha tail weight and the alpha tail weight is just going to be um the amount of the demand distribution uh to the right of a particular tail um so this alpha tail weight is going to be decreasing in alpha um all the weight is to the right of zero and none of the weight is to the right of infinity um and we're going to think about these alpha tailweights as quantifying upside demand potential that you know you have a total mass of one of demand or sorry you have a total demand of one but that demand can either be kind of concentrated around the mean or it can be kind of spread out and we think about the higher these alpha tailweights as being that's sort of the more spread out demand is and the more likely demand is to be very high or very low. So our first result is about products that we're going to call experimental these are products that have high upside demand potential and the former result is that there's going to exist some threshold demand state alpha bar such that if five alpha bar so this is the alpha right tail weighted alpha bar is sufficiently close to one so you have a lot of weight in this right tail then the regulation is going to stimulate innovation and so we think about this right tail weight as being large for experimental products that have significant upside demand potential and I'm going to skip over the details of this example so I can get to some other things that I want but this example is just a very simple two-point distribution that gives a case in which you have high upside demand potential and the key point here is that you either have very bad demand that's close to zero or you have very high demand that has that puts a little bit of weight on a large demand state and so this kind of polarization of demand produces high upside demand potential and in these sorts of markets the insider invitation ban is going to be good for innovation. Now the converse result for a very different sorts of products is that if the alpha tail weight at one so this is the weight of the distribution to the right of the mean is sufficiently close to zero then the regulation is going to reduce innovation and so we think about the upside potential demand potential is being small in markets for incremental products where you understand pretty well how the demand for that product is going to shape out and there isn't going to be a lot left to learn about demand and so the demand distribution is going to be kind of concentrated close to one and so for these sorts of products the regulation is actually going to hurt the amount of product entry under these platforms. Okay so let me skip that this example is just showing with the two more distribution as you move the points of the distribution close to one then indeed you get a situation where you have low upside potential and where you end up getting an incremental product where the ban is going to hurt innovation. Okay so that's kind of a those are our main results and that just gives a sense of some taxonomy of what's the linkage between the distribution of demand for new products when there's uncertainty about it and the impact of a regulation a very blunt regulation where you completely ban insider imitation and now in the last nine minutes of the talk I'm going to tell you about a couple alternative interventions so the first alternative is going to be a more nuanced intervention where you don't just ban data usage outright but you live in it in a particular way that I think is suggested by the name patent. So this is just by kind of a clear analogy with classic patent law which tries to balance the concerns for innovation and kind of the gains from competition by providing limited protection for new products and so analogously we can ask in our setting could the regulator do better than an insider imitation ban by providing innovators a limited data patent and so formally what we're going to allow the regulator to do is to ban insider imitation for a length of time TD after the introduction of a new product following which the patent expires and the platform can now use that demand data if it wishes to target introduction of any product. So here's the the result about how the platform is going to respond to the data patent as a function of its length so not surprisingly if the data patent is short then the platform is going to weigh out the data patent and is going to you know wait until it expires and then it's going to look at the demand data and decide whether to imitate based on the demand data but if the data patent is long then the platform isn't willing to wait until the patent expires and it just goes ahead and imitates prior to that time without waiting for the demand data and in terms of what happens to optimal innovation the key result here is that the optimal innovation threshold is going to drop discontinuously at this threshold time and also for times for for data patents that are longer than t-bar the patent is going to be a functionally identical to a complete ban the platform is going to move ahead of the expiry of the patent that means that imitation is just happening unconditional the demand state and that's going to look exactly the same as if the ban lasted forever and so what this result tells us is that you can do better in terms of stimulating innovation than a data ban even if the data ban is useful that it's even more useful to kind of calibrate a data patent so that you're just prior to this threshold so that the platform waits until the ban expires in order to use this data now of course that only tells us that a data patent can do better than a data ban you know it doesn't therefore tell us that some regulation is called for in this market and so this next slide is going to say that indeed it also expands the scope of markets in which regulation is actually going to be effective so here we're going to show that under some conditions data patents can be useful even when an insider imitation ban is not in when you had an incremental product market where the the regulation would have hurt the amount of innovation so the long the short of it is that when F satisfies a hazard rate regularity condition which i'm not going to show you because it's a little complex to describe but very roughly says that the hazard rate shouldn't be too large in in some kind of endogenous way then the amount of imitation sorry this should actually say innovation not imitation the amount of innovation that's induced by the regulation is actually maximized at this threshold time and while i won't tell you the condition what i will tell you is that all distributions in these categories are all going to satisfy this regularity condition so log normal uniform Pareto and gamma distributions are all regular in fact it was hard for us not to find distributions that don't satisfy it and basically those distributions have to look like distributions with atoms they have to have hazard rates that really explode at certain points and so just the kind of point here is that not that a regular regulation is always called for you can construct counter examples where no regulation is optimal those most easily those involved distributions with atoms but the point here is that there's much broader scope for applicability of data patents than there is for just an hour at data ban um i also want to very briefly mention one other kind of social welfare aspect which i haven't been mentioning so far in the talk but which is very relevant for data patents so one point that i mentioned but didn't spend much time on was that the short data patents they induced the platform to delay imitation intelligent access to demand data and a corollary of that is that there's this welfare boost from tailoring imitation so by tailoring imitation that's not just good for the platform's conversion rate of profits it's also good for social welfare because it eliminates this wasteful expenditure of fixed cost in low-demand states and so there's sort of a welfare boost that is granted above and beyond the change in in the innovation level so kind of taking given the amount of innovation when you get the platform to tailor to the demand data that's good for welfare and so a good data patent design is actually going to kill two birds with one stone it preserves the welfare advantages of data usage by convincing the platform to wait until the patent expires to tailor to demand while also minimizing the negative effects of this imitation on the amount of innovation on the platform so we think about data patents if the regulator has some ability to gather data and tailor it to individual product markets as being potentially a much more powerful tool than just an outright Chinese wall and then I just want to briefly mention one alternative intervention which is often discussed which is just divestment so whenever there's kind of you know popular critical discussion of vertically integrated platform practices you know forced divestment is often brought up as a possibility and you know some politicians have been very vocal about this like I think Elizabeth Warren in particular in the United States has been vocal for divestment the idea of divestment is just that you need to break up this platform into separate marketplace and product companies so that they can't coordinate in any way with each other and of course there might be good reasons to do that to solve other abusive practices like self-preferencing or or kind of a you know rigging of the search results or something which we're not going to touch in this paper but we can just ask a straightforward question of how divestment compares to data policy or to data privacy as a tool for as a kind of tool for regulating the usage of data now certainly divestment does do the same thing as a Chinese wall in the sense that now they're separate companies and so you can it becomes even easier to make sure that they don't share data with each other but it also just changes the nature of the problem that the product division faces because now it doesn't collect fees from the sale of products on the platform and so it's thinking about what products are worth imitating is going to really change versus the case where the there's a single platform that's making integrated decisions about the two divisions and so our result here is that divestment is going to reduce innovation compared both to an unregulated benchmark and even compared to an insider imitation ban so if your goal is just to stimulate innovation because of these concerns about you data usage divestment is a really bad solution for that particular problem and the reasoning here is that there are really two different forces that both turn out to point in the same direction so the first force is kind of a direct one that the divested product division now doesn't get any fees from third party products listed on the platform and so it just becomes a more aggressive competitor it's more willing to trade off less entry for more profits from the the products that do enter but there's an additional force which is interesting and subtle which is that basically these marketplace fees are kind of complementary to data privacy so it turns out that if you take the fees the marketplace gets gets from products to zero then all products become incremental so we showed you know one cut of the problem which is that you can think about experimentation versus experimental versus incremental products as being a function of the demand distribution but of course there also going to be a function of other product market variables which we left fixed and it turns out that if you take the monopoly profits of the platform to zero then every product is going to end up being incremental and so if you then bar access to data that has a further effect of further harming innovation and so you really get this kind of double whammy of negative effects from divestment in fact it's sort of exactly the opposite of what you might have been hoping to get which was that the data privacy you were getting from divestment was hopefully going to kind of help stimulate innovation on the platform and so this is just kind of you know highlighting that that directly kind of behavioral regulation of usage of data may be a lot better than structural regulation of you know the integration of these two services in one company for purposes of affecting innovation on these marketplaces. Okay so let me conclude so I don't go over time so as we've shown data usage for vertically integrated platforms has ambiguous implications for innovation because the platform at least partly internalizes the chilling effect of its own imitation on the entry entrepreneurs and so the the reaction to a regulation is going to depend critically on the shape of demand in particular this upside demand potential that we've identified as a key force and then what we've also shown is that insider imitation bans can be effective at improving the situation depending on the shape of demand and that they're an attractable alternative to blunt instruments like divestment they maintain this beneficent interest in collecting fees from new product introduction and you can actually refine them to do even better by creating limited data patents which eventually expire so I think I'm now a minute over time so thank you all for your forbearance and for your attention. Excellent Eric thank you very much indeed you're exactly on time we started one minute late so that's great so then I asked Justin for his discussion and then I had already collected a couple of questions for the Q&A. Would you like me to stop sharing my screen? Yes please I think then we can see yeah more screens of the people. Okay so let me just start with a big picture I think this is actually and I'm not always the type of discussion that says this so I really mean it when I say it I think this is actually a very important paper and it emphasizes that innovation is inherently a complex phenomenon and then we need to think a little more carefully before we jump into certain regulatory regimes as I'm sure you you gathered from the talk the essence here is that we may care about innovation measured or categorized in different ways so it's not just you know like a homogeneous product so to speak there's lots of ways of thinking about you know what he thinks what you know they think about here is of course the riskiness of the underlying innovation but of course you could also think about you know other types of innovation that maybe are less appropriate here like you know how easy is it to build on a past innovation so you know obviously innovation is a big topic and I think this is actually a really nice contribution okay and as hopefully you gathered the main argument is that you know if you restrict the ability of the platform to utilize demand data and it's vertical integration decisions then you actually in some cases can harm innovation rather than helping it okay what are the key ingredients first of all obviously uncertainty and second the commitment issue okay actually there's two types of uncertainty here one is revolving the innovation cost of the entrepreneur and the other is the demand state okay the commitment issue I just want to say I mean yeah obviously this is important I mean I think there was a question about that already I think it's quite clear that if you if you if you get rid of that assumption you can you know get different results in fact there's a paper that Andre Julian and and Todd have on that sort of same same comment or same same issue okay but at the same time I mean it's I think it's always appropriate to think about the commitment benchmark and I think it's pretty reasonable that a large firm like Amazon potentially would be able to establish some type of commitment okay so I personally don't have any problems with that but they are key key ingredients now I think the issue I've seen this paper several times and it always generates a lot of well what what about this you know what about that type of questions and you know I have a list here I'll I'll touch on you know there's all sorts of you know questions or concerns or potentially objections you might have to you know the particular assumptions they make sometimes with the paper I think that's really problematic because it suggests like that the issue they're looking at is on interesting or on stable in some way here I just think it actually suggests the opposite that this is actually a really important issue that we don't know very much about and yeah sure there are a bunch of questions like well you know what about this what about that I think those are important I think they those suggest and again I don't always say stuff like this when I discuss papers I actually think that there should be a lot of follow-up work that seeks to you know answer some of these questions so just you know just to give you a you know a flavor I'm sure there'll be others and the discussion or rather the after chat I'm not suggesting the author should do any of this by the way just to show that there's a lot going on lots you could think about you know what if the platform can optimize its fees maybe it doesn't matter in the linear model I wasn't quite sure but you know what if it can optimize its fees not just fixed fees and influence things at them you know at the margin but also kind of per unit or revenue sharing fees and influence extraction and say the higher demand states more broadly what if we forget about vertical integration but we think about the platform selling data between different types of firms so maybe there's already three firms in the market they're all generating some sort of data and the platform can now sell this data back to the firm so this actually happens you know Alibaba is kind of a leader and kind of marketplace intelligence and selling that that's a related very important question and then the other main striving to stay under five minutes here the other main issue I think this one is actually pretty interesting again not saying it should be done on this paper another assumption is that the there's sort of symmetric uncertainty between the entrepreneur and the platform regarding the demand state so they're equally uninformed about what demand will be I think it's pretty reasonable that the entrepreneur might be better informed I don't know if that matters for the results but I think you know it would be interesting to think about that I actually have kind of two final thoughts here I think are more important in a big picture level the first was not in the current paper I was thrilled to see that Eric talked about it which is okay fine so there's a problem with you know a simple type of data regulation what are better ways to regulate and it looks like Eric and his co-author are now working on that maybe it's data patents maybe it's something else but I think it's one thing to say okay there's a problem with this regulation but it's quite another to say hey here's a different solution a better solution and then the last comment I have I've mentioned this to Eric before so this is just kind of a more you know just as part of the discussion here a really interesting aspect about online marketplaces is that they're permissionless so if I want to go sell something on Amazon I don't need to be the manufacturer of that product I can go buy that product and then resell it on Amazon and in fact Amazon encourages this here I think it's important because you might have thought well maybe competition will solve this problem you know discipline platforms not you know cheat so much but it's not at all it's not all clear that that would happen and I guess I won't go into all the details just because I'm over the time but if you think about a world where as I said anyone can come and resell a product then it becomes hard to use market competition between platforms to discipline this type of behavior so again just in closing I think it's a really nice paper lots of questions concerns sure but I think in a good way here thanks