 The format is the usual. In a sense, we're going to have 40 minutes of talk and then five minutes of discussion by Daniele Condorelli from Boardwick University and then 15 minutes of informal chat. Thank you and Leo and thank you for the invitation today. It's great to be here and I'm happy to present my paper called Equivalence and Business Models for Information and Intermediaries. Can everyone hear me good? Okay, so now let me start with some motivation and a little bit of background story of what this paper is going to be about. For the past one or two decades, maybe, platform economics has become popular. That's probably part of the reason why we're having this seminar today. A specific feature of platforms is that they help sellers or producers of a product displaying their goods to potential buyers. Classical examples include Amazon or eBay if you think about the platforms in the internet age or you can think about platforms in a more traditional format such as 2VC which is a broadcasting channel that allows sellers to go on the channel and display their products and make advertising. One important thing about these platforms is that they are able to provide information about the products to the consumers which is otherwise difficult for the consumers to acquire or for the sellers to provide by themselves. If you go on Amazon as a consumer, you can easily see the picture of the product. You can easily see the review. You can see all sorts of details of a product which is otherwise difficult for you to figure out if you don't go on Amazon and just interact with the seller remotely. Okay, and these platforms have this technology and therefore they can profit from it. In other words, this paper is going to study an economic interaction where the platforms have the information technology and that can help the sellers inform the buyers about the product and plus that there are many ways for these platforms to operate. An Iranian example throughout the paper is going to be Amazon and Amazon operates under two major business models. Why is the so-called Amazon vendor central? Where Amazon acts as a retailer or a middleman who buys the product first from producers or sellers and then sells the product directly using its own platform or website to the consumers and in the meantime providing whatever information that's suitable for for itself. On the other hand, the seller central business model is the one in which Amazon acts just as a third party who only provides the space or the website or just a place for sellers to display their products and help the sellers provide information and has nothing to do with how the transaction is made. In particular, in the seller central model, Amazon doesn't dictate market price or the product prices. It is the seller who eventually sets the product price and because there are many business models to operate, a natural question arise which business model is better, which business model is more profitable for a platform, which business model is better for consumers, which business model is better for sellers, and which business model is better for the economy. So in this paper, I'm going to study this kind of question and the main result of the paper is actually that well, in some cases, business models don't actually matter. I'm going to, the main result is going to be that all the business models are going to be outcome equivalent if and only if some virtual markups are large enough. And I'll be more precise about what this virtual markups mean. Basically, you can think of virtual markups as a decreasing function of the seller's information rent or an increasing function of the demand elasticity of the market. Okay, so that's the motivation. And now I'm going to jump into the model. The model is relatively straightforward. There's one product, one buyer, one seller, and one intermediary. The buyer has a unit demand and a value v, where v belongs to a compact interval, cap v, that's non-negative. And let me let this d not of p denote the distribution of that value so that d not of p is the probability that a value p is above the value v is above a price p. So we can think of d not also as the demand function induced by this probability distribution. On the other hand, the seller has marginal cost c that belongs to another compact interval. It follows the common prior g where g has a density little g that's everywhere positive. Now, let me let f g denote the usual virtual cost function as it calls the plus and information rate. Okay, and let me assume for this talk only, this is just for simplicity, that f g is increasing so that g is regular and d not induces a decreasing margin of revenue curve. In other words, d not is regular in the minor study as well. Okay, so that's the primitive of the model, not going to the information. The buyer doesn't necessarily know what v is exactly. The buyer has to learn about v through a signal. A signal is formally a transition probability that maps from true value to a distribution over values. In other words, a signal is a Blackwell experiment that maps from the state space, which in this model, the buyer's true value to a distribution over signal realizations. Now, because the buyer eventually is only going to make a binary decision of whether or not to buy the good, it is without a loss to assume that the signal space equals the state space, which is v here. Some examples, we can have full information fully revealing the buyer's true value to the buyer so that for every v, pi of v is a direct measure that puts probability one on v. Or we can have a signal that's not informative at all. So, pi of v is a direct measure that puts probability one on the expected value of v for all v. Or we can have partial disclosure so that for any v, if v is below a threshold v star, pi of v is a direct measure that puts probability one on v lower bar. If v is above a threshold v star, then pi is a direct measure that puts probability one on another variable, v upper bar. These are just examples. I'm allowing for all kinds of possible signals, which is a transition probability of this form. Now, given a signal and a true value v, the buyer is going to observe privately the signal realization that is strong from pi of v. And then the buyer is going to update the base rule before he makes a purchase and decision. Now, the price is going to be determined by the mechanism and business models, which I'll explain later. But at the time when the buyer makes a purchase and decision, he already updated after receiving this private signal realization. So, for a buyer, he's going to decide whether or not to buy the product at some price based on his posterior. Now, a good feature, a nice feature of this model is that the buyer has unit demand, and therefore only the interim expected value matters for all the information that's relevant. So, we only have to keep track of the buyer's interim expected value, which is here. Given a signal realization as the buyer will form expected value, that's the expected value of v conditional as. Okay, let me let d pi denote the distribution of this interim expected value. This interim expected value is itself a random variable, because as itself is strong from a distribution pi of v. So, there is a marginal distribution of this random variable. And I'm denoting this marginal distribution as d pi. Of course, we know from the law of iterated expectation, the expectation of this interim expected value equals to the extended expected value. This implies that d pi is a mean preserving contraction of the prior d naught. Conversely, that's also true. For any mean preserving contraction d of d naught, there exists a signal pi such that they induce the distribution of interim expected value equals to d. Okay, and this observation means that we can just represent the entire set of signal, which is quite large here, the set of transition probabilities by just a collection of mean preserving contractions of the prior. I'll formally write that down before that and move this up, formally writing down the collection of mean preserving contractions of the prior. So, script d naught is the collection of non-increasing upper semi-continuous functions such that the following integral inequality holds. At all p, that's now negative, and with equality at p equals to zero. Now, this is just a classical Roth style, Stiglitz style representation of mean preserving contraction condition. And it says that any function, non-increasing function d satisfying this integral inequality is a mean preserving contraction of d naught. So, now I can replace this long sentence by just the notation script d naught, and I can replace the definition of a signal by just sum d and d naught. Now, this representation is convenient because remember, I started with tracing the marginal distribution of the interim expected value, and then representing entire signal by just that distribution. This means that for any signal now called d in script d naught, it summarizes the distribution of the buyer's interim expected value. And because the way I'm defining d, d of p is the probability that the buyer has an interim expected value that's above price p. So, a signal is also itself a demand of the buyer under that signal. So, now I'm using the buyer's demand to represent the buyer's signal. And where does the signal come from? Well, it comes from the intermediary. So, now let me turn to the intermediary side of the model. The intermediary can operate under different business models as motivated in the introduction. And for the majority of this talk, I'm going to focus on two major business models. And in the end, I can talk a little bit about how to generalize it and how to think of this result that involves only two business models. The first business model I'm going to consider is what I call the strong contracting model. And that corresponds to Amazon's vendor central model that we just saw in the introduction. In this business model, the intermediary is able to contract with the seller on every aspect of the market, including the information that's going to be provided to the buyer, the price that's going to be charged to the buyer, and the amount of money that the intermediary that the seller would have to pay to the intermediary in exchange of the service. So, the revelation principle applies here and therefore a mechanism, we can just look at the direct mechanism that asks the seller to report her cost so that given each report, say C, a mechanism specifies DFC, which is a signal that's going to be provided to the buyer. Gamma of C, which is the distribution over price from which the seller would have to draw when selling the goods to the buyer. And TALFC, which is the amount of money that the seller would have to pay to the intermediary. And as usual, we can write down the IC and IR constraint in this environment. IC means that the mechanism would induce truthful report and IR means that every truthfully reporting seller or at least her outside option, which is zero in this case. To read this IC condition, let's start from the right hand side. The right hand side of this inequality is the net profit of a seller whose true cost is C and when she reports C prime. Because she would have to pay TALFC prime to the intermediary and the mechanism is going to generate a signal DFC prime and the seller would have to sell the product according to the distribution of price Gamma of C prime. But then this inequality says that the seller wouldn't want to miss report and send for the IR constraint. So that's the strong contracting model. Let me keep the mechanism here as a reminder. Guy, quick question. Are you assuming throughout that there is full commitment so Amazon can commit to a mechanism and is this reasonable for all sellers operating through Amazon? This is a question from Luis. Okay. Yeah. In some sense, I am assuming throughout that the platform has commitment power. And as we can see in just a minute, the main difference between two major business models I'm considering is exactly the degree of commitment power. And well, just as a preview, in another business model, the platform or the intermediary would not have the ability to contract on price, but would have the ability to contract on information. So from that aspect throughout the paper, the intermediary is assumed to have commitment power on information. And that's a simplifying assumption, of course. And that's a way to model communication and information transmission. Of course, it's not the most reasonable or it's not the only way to model it. That's one way. But the main focus of this paper is to look at different level of commitment power in terms of how influential the intermediary is in the product market and in particular over prices. What that helps. Yes. I guess we can postpone the discussion of what happens when sellers of market power for later on when they're strong. Okay. Yeah. Yeah. Okay. Thank you. And okay. So one thing I'd like to remark is that this strong contracting model is exactly the retail model that I just mentioned for the Amazon vendor central. Because although Amazon vendor central acts as a retailer, in a strong contracting model, the intermediary is able to contract on everything. So the intermediary could just contract on the prices and information that it would have used when it already got all the products from the seller and acts as a retailer. So this strong contracting model in the background is going to be an approximation of Amazon's vendor central. That's mentioned in the introduction. Now the second business model is going to respond to the Amazon seller central. And this business model as just as I was just saying, the intermediary is not able to contract on prices and can only contract on signals. So a mechanism here again asks the seller to report her cost. But given it's reported the cost, a mechanism can only specify the signal that's going to be provided to the buyer and the amount of transfer that the seller will have to pay to the intermediary. IC and IR constraints are as usual. IC means that the seller wouldn't want to miss report and IR means that every truthfully reporting seller earns at least zero. Now there's one thing interesting in this IC condition. If we just look at the right hand side, that's the net profit of a seller whose true cost is C and when she reports C-prime. Because she would have to pay Tile of C-prime and the missed report is going to generate a demand D of C-prime via signal. But at the end, it's the seller who gets to choose the price. So the seller after signing a contract with the intermediary would have to sell, would have to solve a profit maximization by herself. And that profit maximization involves the seller's true cost as opposed to the missed reported cost. And that's going to be a key ingredient that creates some economic tension that we have to deal with later on. Okay, so that's the weak contracting model and intermediary under both business models is just going to maximize the expected revenue. All right, so that's basically the model. And a few words on what's going on here is that first of all, the strong contracting model, as I just pointed out, is basically modeling the Amazon vendor sent for retail model. And it turns out that it's just a standard one-dimensional classical screening problem, just as in Musa Rosen or Myerson or other classical papers. And the reason is that as a retailer, the only private information that matters is the seller's production cost. And the only allocation that matters is the quantity purchased from the seller. Of course, there's going to be a continuation gain for that intermediary after purchasing the good. But that continuation gain is a more of a standard persuasion or information design problem. So in a strong contracting model, the ability to contract on price simplifies the problem or, in other words, allows us to separate the problem into a classical one-dimensional mechanism design problem and a classical information design problem. The same thing is not true in the weak contracting model, because in the weak contracting model, the seller gets to choose the price after a signal is determined by the contract. In other words, the weak contracting model is a screening problem with an exposed non-contractable action, which is the price charged by the seller here. But in any case, a strong contracting model is mathematically a relaxed problem of the weak contracting model, because there are just fewer constraints. We don't have to worry about prices in the strong contracting model, but we do have to worry about that in the weak contracting model. Okay, so that's the model. And now I'm going to state my main result before going into the details of the ideas of the proof, unless there are any questions. Okay, so before stating the result, let me define formally what I mean by outcome equivalence. I'll say that two business models are outcome equivalent if, under any optimal mechanism of these two business models, the following understanding, the intermediary's revenue, expected revenue, the seller's interim profit, so that's a function of the realized cost, the buyer's expected surplus, and the X post allocation of the good. Okay, so basically everything. And let me let R not denote the usual revenue function as a function of quantity induced by the demand of D naught, and define this mu function as the following. So that mu of Q is the inverse of the virtual cost function evaluated at the inverse demand minus the marginal revenue at Q. Okay, I'm calling this term the virtual markup at Q. The reason is, if we first for a moment suppose that there's no information rent, so that the virtual cost equals to the actual actual cost, then the inverse of the virtual cost function equals to the identity function. In that case, mu of Q becomes just an inverse demand minus marginal cost. And we know that this term is called the markup, because if you remember this picture, here's the inverse demand, here's the marginal revenue, and if Q is here, we often said that the length of this segment is the markup charged by the monopoly whose cost is such that Q is an optimal quantity. Now adding back the information rent, we have to readjust what this demand looks like, and it's going to shift down, but in any case, the difference over here is going to be interpreted as the virtual markup, because now we're adjusting the location of the inverse demand via this virtual cost function. Okay, now let me keep this definition over here as a reminder, and state my main result. The main result is that there exists an increasing functional lambda, with lambda zero being greater than or equal to zero, such that the strong contracting model and the weak contracting model are outcome equivalent, if and only if lambda mu, where mu is the virtual markup, is greater than or equal to zero. Now this means that if the virtual markup is large enough, then we'll have outcome equivalent. And how large? Well, remember in the benchmark case where there's no information rent, mu of Q is always no negative, right, because the markup charged by a monopoly is always no negative. So when there's no information rent, the two business model is always equivalent. And lambda being increasing means that when the virtual cost function creates a distortion that's small enough, we'll still have equivalence, or alternatively, if we have a demand function that induces a small enough or a low enough marginal cost, marginal revenue curve, which means that the demand function is more elastic, then the virtual markup is going to be large, monotonicity of lambda means that the outcome equivalent is going to hope. Okay, so that's the main result. And now I'm going to spend some time to basically sketch the main idea of the proof. Because of time, I won't be able to give the full proof today. And in particular, the functional lambda, I have a closed form for it, but it involves a little bit of derivation. So I'm not going to do that today. And instead, what I'm going to do is that I will impose a sufficient condition for lambda mu being greater than zero, and illustrate how in that situation, can we construct mechanisms in two business models that induce the same optimal revenue. Okay, so now let me go into the proof. First of all, for any C and for any D, let me denote PD of C as the optimal price of the seller, whose cost is C when facing the demand D. If there are multiple optimal prices, I'm just pick the largest one. And also remember, the speed G is defined as the virtual cost, which is cost plus information rent. Let me keep that here as a reminder. The first step of the proof is the usual revenue equivalence formula. It says that on the NEIC mechanism in both business models, the expected revenue can be written as a function of the allocation in that mechanism up to a constant. Now, this is just a necessary condition for IC, and it's far away from sufficiency, partly because one, even in standard one-dimensional mechanism design problem, a revenue equivalence formula is not sufficient for IC. We have to have a monotonicity condition. And that's true for the strong contracting model because, as I said, the strong contracting model turns out to be a one-dimensional screening problem. So in the strong contracting model, this revenue equivalence formula plus the monotonicity condition of the quantity sold as a function of cost is going to be necessary and sufficient for IC. But that's not true in the weak contracting model. The weak contracting model involves exposed non-contractable actions. And in particular, there is some possibility of double deviation of the following four. A seller with true cost C may want to misreport and try to pretend to be cost C prime, and that misreport is going to generate a demand C prime. And yet, the seller has true cost C. So the seller is going to price optimally according to the true cost rather than the misreported cost C prime. And this double deviation concern means that a simple monotonicity of the quantity over here is not going to be sufficient for IC. And some more complicated condition for those who know this literature, it's going to be some form of integral monotonicity condition that would be required there. But in any case, that's a complicated condition. And I'm not going to be able to get into the detail for today. So today, I'm going to focus on the necessary condition. And I'll tell you what the optimal mechanism looks like. And if there are interests in the end, I can talk about how to verify that the mechanism is indeed IC. Okay. One thing that I want to notice is that under the weak contracting model, if you look at the integrand of this expected revenue term, it summarizes the key tension in this model quite well. But there are actually two layers of distortions going on here. The first layer, well, first, let's notice that the integrand is the expected profit of the seller whose cost is replaced by the virtual cost, and yet is still pricing optimally according to her true cost. Now, this highlights two layers of distortion. One is the usual screening slash address selection distortion because we replace the cost by virtual cost, meaning that the intermediary faces a seller who has private information about cost, so it would have to pay some information rent to the seller that effectively increases the intermediaries margin of cost compared to the sellers. So that's the first layer. But the second layer over here is that at the end of the day, in the weak contracting model, it's the seller who gets to choose what price to charge. So the price has always to be optimal with respect to the true cost, as opposed to the marginal cost. So even if the intermediary has effectively a higher marginal cost, it cannot just force the seller to charge a price that reflects that marginal cost. The seller would always charge a price that reflects her true marginal cost, and that highlights the second layer more of a moral hazard tension of this weak contracting model, and that's going to be the main problem that we're going to solve here. Now, because of that, the weak contracting model is not a standard mechanism design slash screening problem. So there's no off-the-shelf tools for us to solve that. The way I'm going to do that is that I will guess an upper bound for the expected revenue. I kept the expected revenue over here as a reminder, because I'm going to erase this soon. I'll guess an upper bound for the expected revenue, and then I'll find a mechanism in the weak contracting model that gives the same amount of revenue as that upper bound. Now, let me try to derive that upper bound. First of all, remember that this was stated in the previous lemma. Under the strong contracting model, under NEIC mechanism, the expected revenue looks like this. Now, I dropped the constant because by DI, our constraint means that the constant has to be zero. But under the strong contracting model, the expected revenue for NEIC mechanism looks like this. Now, whatever this thing is, if we look at the integrand, that's the profit of the seller whose cost is replaced by the virtual cost when facing the demand D and when drawing a price from distribution gamma. Whatever that is, it's going to be bounded from above by the total surplus of the economy, where I replace the cost by the virtual cost, because the seller's profit is going to be bounded from above by the total surplus of the economy. All right? And I'm going to separate this integral. So rewrite that in the following way. And I'm going to call this thing R star. Now, one way to think of R star is the following outcome. We have something minus virtual cost times a probability of trade. So we can think of this outcome as inducing sell if and only if the buyer's value is above the virtual cost, meaning that we're selling to all the buyers with true value that's above the virtual cost. And we're going to sell the buyer at the price that equals to the conditional expected value, conditional value being above the virtual cost. Okay? Now, what I'm going to do now is to find a mechanism in the weak contracting model that gives expected revenue that equals to R star. That's going to solve the problem because remember, the strong contracting model is a relaxed problem of the weak contracting model. So the optimal revenue under the weak contracting model is bounded from above by the optimal revenue under the strong contracting model, which is in turn bounded from above by R star. So if we can find some mechanism in the weak contracting model that gives expected revenue R star, then that means that this inequality chain must be all equalities. Okay? So that's what I'm going to do. Going back to the weak contracting model, before going back to the weak contracting model, notice that if we were in a strong contracting model, then we're pretty much done. We kind of know how to achieve this R star by sort of the standard persuasion argument or the result in a classical paper by Anderson Ronell. Because in the strong contracting model, the intermediary is able to choose price and information at the same time. So to achieve this outcome, the intermediary can simply just offer a binary signal that tells the buyer whether or not his value is above or below the virtual cost. And then force the seller to charge a price that equals to the interim expected value, conditional value being above the virtual cost. And that way, we can induce the desired outcome. But that's not true for the weak contracting model because price is not contractible. And in particular, it may not be to the seller's interest to sell the product at a price that equals to the interim expected value, conditional value being above the virtual cost. When the seller's true cost is C, as opposed to the virtual cost VG of C. So in order to induce the seller to price in that way, we have to do something else. In any case, the goal here is to find an ICIR mechanism, V star, tau star, in the weak contracting model, such that for every seller, she would sell to the buyer if and only if the buyer's true value is above the virtual cost, and she would sell to the buyer at price that equals to the expected value of V, conditional V being above the virtual cost. So now I'm going to construct this mechanism, V star, tau star. Before that, here's the sufficient condition that I mentioned, that I'll impose for this top, just as a simplifying condition. I'm going to assume that VG is pointwise below the optimal price under the prior. And this would imply that this lambda mu condition in the statement of the theorem is always non-negative. Okay. And now I'm going to construct the mechanism. The transfer is going to be pinned down by the revenue equivalence formula. So I'll just focus on constructing a mapping from the report cost to and being preserving contraction of the prior, which I will call D star. Let me draw the prior over here first. For any report C, the C is here, and the virtual cost is here, define V of C as the targeted price, which is the expected value, conditional value being above the virtual cost, that it's here. And now I'm going to draw the mean preserving contraction, D star of C, as the following, and I'll color that in blue. And it's actually quite simple. For every value that's below the virtual cost, we're going to keep that distribution the same as the prior. And for every value that's above the virtual cost, I'm simply just going to concentrate all the probability weights at V of C. And that's D star. And here's a formal definition of it, in case someone's wondering. But one important feature of this D star is that because of the definition of V of C, we know that the area of these two regions are going to be the same. And this would imply that D star is indeed a mean preserving contraction of D zero. Furthermore, it turns out that is indeed optimal for the seller with cost C to charge a price at V of C. Because, first of all, remember, I had this assumption that the optimal price of the priors above the virtual cost, but the optimal price is going to be somewhere here. And second, remember, I had this assumption where the prior demand induces a decreasing marginal revenue curve. And that when translated to profit function as a function of price would mean that the profit function is single peak like this. And of course, the peak is going to be at here, which is to the right of the virtual cost by this assumption. And this means that under the prior, charging any prices below the virtual cost is going to give less of a profit than charging a price and charging a price at the virtual cost. But that's not just under the prior. D star looks exactly the same as the prior when the value is below the virtual cost. So the profit function under D star, which is in blue here, is going to look exactly the same as that of the prior, which means charging any prices below the virtual cost is going to give less of a profit than charging a price at the virtual cost under D star. But virtual cost is not an optimal price either. Because we can see from this demand, we can always increase price without changing quantity until we get this point V. In terms of profit function, it means that the profit function is going to keep increasing linearly until we get to the point V of C and then drop to zero. Now from this profit function, we can see that the optimal price of the seller with cost C is indeed V of C. Now, why is this useful? Well, it means that under D star of C, the buyer is going to buy if and only if value is above V and they're going to buy exactly at price V of C. That's the desired outcome we want to attain. So let me keep this D star and move it up. I use the revenue equivalence formula to back out the transfer. Here's the optimal mechanism and that mechanism attains revenue R star. Now remember, the proof is not done yet because I have to check that this mechanism is indeed incentive-compatible. I might not be able to go through that today, but if there are interest in doing the Q&A session, I'll be happy to talk about that. Basically, what happens here is that there's a screening cost which effectively boosts the intermediary's marginal cost so that it's higher. So the intermediary would prefer a higher price than the seller does. Under a strong contracting model, the intermediary can just force the seller to charge whatever price that it desires. But that's not true under the weak contracting model. Under the weak contracting model, the intermediary has to incentivize the seller using the information provided to the buyer. And under the sufficient condition that I gave, the derivation that we just saw means that by fully revealing the value V to the buyer, when V is below the virtual cost, that's going to give the right incentive to incentivize the seller to price in the right way. And in general, whenever the virtual markups are large enough, there are going to be enough of rooms for us to play around with information in a similar way as we just saw so that the intermediary can incentivize the seller to price in the desired way. Now one last thing before concluding is that one implication of this outcome equivalence result is that for the intermediaries, at least in this model, they have no strict preference over a particular business model as long as the virtual markups are large enough. An interesting anecdote is that throughout the past decade, the seller central of Amazon has grown from 26% to 52%. So basically now Amazon is half and half between two business models. And one last word about business models in between, the two business models, the strong contracting and weak contracting models are quite strong, quite strong of an assumption. One allows full contracting ability, one allows none of the contracting ability in the product market. But equivalence result means that any other business models in between are also going to be equivalent when the two extremes are equivalent. Okay, and that's basically I have to say for today. Thank you. Thank you very much. So let's hear from Daniele Condorelli from the University of Warwick who's discussing this paper today. Daniele? Hi, hello everyone and thanks for the invitation. It's been a while I had the mind to look at Kai's work and so this gave me a good opportunity to do it. Okay, so I've been asked to keep it non-mathematical and it's a pleasure I will do that. Kai has been very clear on explaining what mathematics drive the results. So let me try to give you an explanation without hopefully embarrassing myself, trivializing too much. And so I'll keep the perspective of theory as opposed to thinking about platforms and having more of an applied stance. So what this result is about, this is a sort of neutrality results. We know neutrality results are central in economics, neutrality of money, the neutrality of equity debt financing in the Modigliani-Gilder theorem and of course the neutrality of the payment rule in mechanism design or revenue equivalence so-called. So all these equivalence results are important because they allow sense to see that in a certain idealized environment certain specific phenomena are not the cause of in this case the outcomes of our model. Okay and so this allows us to think about I don't know other frictions that might lead to in fact the business model to matter like you would think like a neo-kinesian model could make money no short time if you want or you may take the approach that let's forget about the business model and let's try to think about more important aspects. And this is what a lot of applied theory, applied mechanism design does by saying okay let's focus on the outcome as opposed to then to the payment rule. Okay now of course this paper does not make such a grand claim I would say because it's still it's probably and maybe it requires some thought and I'll say a few words later but but it allows us to in a sense sort of at least what tells me is how to abstract away a little bit from the aspects of information here okay and so I think there's so when you want to think about so the way I think about this is the following there's this seller there's this intermediary that's the buyer the buyer really here plays a very little role in a sense so there are two informant party the seller and the intermediary and what you see is that the intermediary here has a monopolistic power in the sense that if the intermediary do not consent to a contract or whatever sort with the seller then there's no sale okay because the seller has no access to the buyer okay so this monopolistic buy this monopolistic sort of control plus information on the two sides is going to tell me that rents will stay the same okay so the information rent of the seller will have to be there it cannot be abstracted away and so the remaining will stay with the intermediary because of course the buyer is not going to make any money because there's no information here okay so so so in this sense and I think this is natural in a in a equivalence theory you can see into sorry they they they I need to open I'll work with you so so it's natural intuitively that that you would expect this result as in in a sense as in neutrality of money but it's not at all obvious or not how you would how you would now why I said that maybe there's still some work to do so I think one this is this is a potentially larger problem you might think or at least when I think about it the more natural assumption for me is to think that the intermediary is more informed than the seller but no more informed than the buyer often okay and when I think about amazon okay amazon might know about my taste is more than the seller but I will still having some information rent vis-a-vis amazon if I need that product or if I don't need it and so forth so when you think about this situation and then you look at these two models then I think you see that there will be a difference okay because now the value of the information is going to be different to the intermediary than it is to the seller so if the intermediary buys from from the seller in order to resell you will face first a symmetric information problem with the seller okay then once he has bought he will face another information problem a symmetric information problem of selling with the buyer okay so you will have this double marginalization thing here however if the intermediary sells the information to the at the point different problem here in which the seller the buyer is informed but as a symmetric information about the cost of the buyer about the cost of the seller and the seller has now information about the value of the buyer but as the information provided by the intermediary okay so in my conjecture here you would not observe the same equivalency here so I think that that needs to be studied here but this paper in a sense makes progress along one direction and and one point that as I said it might be intuitive but as we have seen from the slide is not at all obvious so well done and I have sent other minor detail come to comment to Kai directly without discussing them here okay thank you thank you thanks Daniele so what we can do is to give Kai the opportunity to maybe react to Daniele's comments and then I have one question holding in the chat and then we can open up the Q&A session Kai do you want to react? Yeah yeah sure okay so perhaps very quickly and thank you Daniele for for the comments that you sent and you mentioned today and I apologize that I lost you for about 30 seconds there so I might send you an email asking clarified questions but based on my understanding I think you are I appreciate the interpretation of of neutrality result and on the broader sense I think one of my agenda is to is to look at that kind of neutrality in similar context this paper doesn't make the grand claim that says that all the business models are neutral but I think you're right pointing out that the equivalence results stated here even though we don't interpret that in the literal sense that Emily doesn't care about business model it allows modellers or economists to think about business models by abstracting away some parts that are that are equivalent and so now we can think more clear about for example how how useful and how important is the cost of being a retailer perhaps inventory cost shipping cost and things like that that's going to be to one order more important than the informational factors because because of the result that we saw in this paper and I agree entirely that it is quite an assumption that the fire doesn't have any private information over here in the model and that may or may not break the equivalence result in general I'm still working on that and that's that's in fact some some work in progress and thank you for pointing that out but one thing I can say is that I have another a related paper which is basically similar to this but now instead of assuming that the buyer doesn't know anything it has to learn things from a signal there the buyer knows everything about the value and the intermediary is going to inform the seller about what the buyer's value is and combining the results of the two papers together we can see that whenever the equivalence result holds here the equivalence result is going to hold there and the amount of revenue is going to be the same so this kind of means that well as a platform you have two kinds of tools one is through informing the buyer through information manipulation in the in the product market by informing the buyer the other is sort of by selling information about the buyer to the product seller and these two are going to be equivalent if the equivalence result holds here and that relies on additional results in my other paper and I'll be happy to talk more about that offline but but thank you very much for the comments they're they're very helpful thank you