Pricing control and marketplace design with Gerry Tsoukalas | a16z crypto research talks

Transcript

thank you everyone for being here um this is actually the start of what we're calling governance week uh at a16z crypto and the research team uh we'll be having a series of talks about governance they're going to get actually increasingly specifically about governance our very first paper is much broader uh and then we'll be with increasing specificity going pretty deep into different governance questions in web 3 and Beyond um without further Ado I'll turn it over to our first Speaker uh Jerry Talis from uh the business school upu thank you Jerry thank you uh Andy it's great to be here uh very grateful for the invitation and there's a reason why it's going to be increasing and the reason I picked the this particular topic is because last year one of my co-authors

was here who was doing uh hardcore voting uh information aggregation stuff that we've done together and so I did want to overlap uh so I figured I'd pick something a little bit broader but I'm hoping to convince you that there are interesting links with web 3 with governance on the application layer right it's not so much poten on the blockchain layer but the the the upper layer so we're going to talk about who controls pricing on platforms very very quickly I know that's it's a relatively short talk so just to motivate this tech companies care about marketplaces and um there's multiple reasons one of the reasons I want to highlight we can just skip most of the quotes here just to the middle one um the reason is that it helps them diffuse the technology okay so let me just

quickly read it says in 2017 every Tech Giant made a big splash about their machine learning as a service platform intended to democratize AI by enabling companies less blessed with technical talent to partake in the latest breakthroughs nice language essentially what they're saying is small and medium-sized Enterprises don't have the resources the capital skills to actually acquire lot of the really cool tools that people are building in Ai and blockchain and all sorts of Technologies including the people in this very room right and so they see marketplaces as a way to diffuse that to these types you know to the broader economy essentially so uh let me zoom in on one and this is a snapshot of macrosoft

azure we're going to get into the model right away and this will help me motivate the model okay so it's fairly simple on the left here you have a list of technologies that you you know a user a buyer might be interested in there's AI machine learning there's blockchain down here and all sorts of uh interesting things once you click on it you get a list of possible vendors that are providing these tools that you can download you can pay for and you could either Implement yourself or you could you know it's part of the package you could actually um get help on the implementation and I know it's kind of hard to see but there are a lot of different products here from a lot of different vendors small small vendors very large vendors vendors that are

known vendors that are not known Etc it's a mix of things and some of these products like this chatbot competes with potentially other chat bots so um they're substitute Goods whereas others are uh complements for instance Nvidia down here is offering an optimization tool that you can then pair with an llm chatbot to make it run better etc etc so you have this this variety so um I want to just take a step back to together thinking about the economics of these platforms it's fairly simple uh what are the some of the things that as a designer you know we might care about so you want to have a healthy system and you want to have a lot of sellers it's one of the

requirements and so you need something like free entry for people to come in you don't want spam obviously but you want some sort of open platform the products themselves typically like I said are differentiated right so they're either substitutes they're competing against each other or the compliments so they can help each other out and when we start thinking about you know from design perspective what do we control what is a what do a platform control not much right okay I'll I'll I'll add a caveat um there are things that could matter like you know user interface design what what font you're using that might affect people marketing budget all sorts of

stuff we're not covering any of that right we're covering matching and pricing which is typically what you look at for platforms but there's not really any matching going on here right the matching algorithm is essentially that search filter that says put in what you want I'll show you you know the sellers basically so we're going to focus on pricing and in this particular platform um the way they monetize is through simple commission contracts really basic it's what most platforms are using they take a percentage anyone guess what the percentage is 30 how much 30 30 30 anyone else just out of curiosity I had no clue

when I started looking at this I mean there is a right answer but it don't worry you if you're wrong it's fine 10 10% okay so they started with 20 it's pretty sizable that was 5 years ago it's down to three now why because there was a little bit of a mini arms race you know Google got in um you know Amazon web services so that drove things down way down okay but they do make some money uh the other interesting part which is the focus of the talk is the decentralized pricing structure which just means you know what we're going to let people price their own products that's it we're not going to get involved and that's what we're going to

focus on um we looked at almost a 100 different platforms out there in various Industries you know all over the place place and roughly speaking you can categorize them in these three types of pricing modes so what I just talked about is this model the decentralized model that's Microsoft Azure uh Google's Marketplace Amazon web services all that stuff and many many others um there's a decent amount of platforms that use the opposite of that model which is the centralized model that says I want to control everything I'm not I don't trust any of my sellers so Uber is notorious for controlling everything in including the matching that we're not going to talk about today but on the pricing side not only do they um decide how much fees

to take from drivers but they also might show different prices to the actual customers okay and they're hoping that customers and drivers don't necessarily talk during the rights um there's other ones open AI announced that it's likely going to be centralized but you know their strategy isn't fully developed yet web 3 probably what people are most interested in in this room is essentially the decentralized model where you remove the platform and you end up with a disintermediated Model A disintermediated decentralized it's a very long let's just call it web 3 there's a long list of startups I'll just focus on one in the interest of time which is Jensen it's a really interesting company the reason I'm

highlighting it is because I really like the way their marketing materials are going after the protocol dis intermediates cloud oligopolists like AWS and does not charge a margin which means a commission right so they essentially allow buyers and sellers to trade to transact for free it's pretty interesting model just a side note if you're interested it doesn't matter for the model but what these guys are doing I think is really really cool they're uh trying to leverage idle compute on everyone's laptops for instance right now you're not really using your laptops you could potentially connect to this Jensen Network and make some money uh so other people can use it to train their uh deep learning models really

interesting um idea okay so there are a bunch of research questions um you know you you might think of other ones as well that we're looking at does the pricing structure actually matter okay so you have centralized decentralized so what right does that actually affect is it a first order question or not um does it affect adoption um and profits and things like that and the second one are there potentially better uh contract structures so this one's not obvious because when I looked at all these platforms without exception they're all using commission contracts outside of web 3 right all the traditional ones whether they're centralized well centralized it can be even more complicated or decentralized it's just

these standard contracts that we all use in many many aspects and does the web3 business model make sense that was a question from Ethan here are the answers why because we might run out of time um so at least I I want to I'm going to give you the whole thing on this on this slide so you you can take away uh something so yes actually it matters the first order thing whether you're doing centralized or decentralized it matters quite a bit there are potentially many reasons there is one reason in the in the paper that we're trying to write that I think is kind of clear and the tradeoff uh I'll just summarize it here that we'll get into it in the model it's a story of two different types of frictions that you that you worry about when you're centralizing sounds like the right thing

to do you get control of pricing you know if you're the platform what you don't observe are people's willingness to work in other words you know concretely you don't observe their costs their opportunity costs their real cost Etc so there is information asymmetry and you know econ 101 that will create some frictions you'll lose a little bit of value in your systemwide because of this friction if you know you're not using um optimal contracts necessarily that's on on the one side if you decentralize now you give okay give everyone control which is the simpler thing to do I mean you can set up a Marketplace pretty easily when there's no matching involved and you give up pricing there's not much to do other

than choose your font I guess and write a little bit of code so when you do that it turns out that it alleviates the information asymmetry problem because people now can set their own prices observing to some extent their costs or their willingness to to join the platform and work what you lose is coordination and you can think of you know potentially two separate models one one of them is everyone is rational all the sellers are rational uh and they're self-interested so they don't take into account what's good for the platform as a whole and I'll show you you know the math behind that you could even think of a model where you know not everyone is rational and automatically you get you know potentially bad outcomes from that type of model but we're going to stick to you

know rational expectations models for for this talk so it's it's a tale of information asymmetry frictions on one mechanism versus coordination frictions when you have the other mechanism okay that's the whole that's the whole paper for the most part it turns out the reason that um you you have you have these two uh trade-offs is because of the commission contract itself which seems innocuous it seems like the right thing to do or the simple thing to do in general but when you're using commission contracts turns out that neither of these things neither of these modes are optimal across all possible types of different markets and competition levels and I'll show you some graphs so what can you do then we take a mechanism design approach and we solve

for the optimal contract which turns out I couldn't really find a name it might there might be in the literature maybe some of you might know it we call it iine contracts in the paper you know it's a tweak on the commission contract I'll show it to you um when we get there and then does the business web 3 business model make sense ignoring the tokon the tokenomics aside let's imagine they don't monetize and they're altruistic that's best case scenario um I'll show you that even in that scenario it can fail when they're using commission contracts okay question yeah um so like in the contract theory literature sometimes you know these pure commissions are Justified on robustness grounds yeah

like I don't know if you've seen like gab car it will be more robust I'll show it to you yeah okay but it's not it's not optimal well it depends on your notion of optimal right so robustness in the sense that it's like worst case optimal over information you may not know right yeah exactly yeah okay but it's not yeah without that constraint it's not globally if you know all the details of the yeah yeah absolutely absolutely um there's a literature review just I know it's a short talk so I'm I'm going to skip uh it's also a little bit outdated the only point I want to make is that there's a growing literature on this as you can imagine um there's old literature but doesn't really answer the questions that we care about for web 3 so there's this new literature that's really looking at

especially at the application layer how do you actually create these Dows uh govern them you know which decision do you you put in a smart contract how do you do it Etc on the model side the uh decision variables I'm trying to keep everything in blue and for those of you who are modelers or theorists in in the room and endogenous quantities are going to be in red and everything else in Black just going to be an exogenous parameter that we um we fix so um this is one of the challenging parts of writing a paper like this there's a lot of different players there's the platform so we're going to assume that they cannot directly observe private costs but they can observe some distribution of costs and they're going to set the contract terms which is just a a simple commission or in the

mechanism design part of the paper it'll be just some general fee that could be anything the buyers I'll come back to it in the next slide they are in Gray here in the picture and the sellers are in blue um the sellers are free to enter they offer differentiated Service uh which essentially means that either they're you know competing a lot because their products are similar or they're not really competing a lot because their products might be very different and they have important asymmetric private costs um that I list out here and then we look at the three pricing modes so in the centralized mode the platform sets the price cannot observe the costs of these of these sellers there's a single price for everyone since it doesn't have any other information I think the right way to

think about this is that you know it sets the price for a given local market at a given moment in time and then it could change that in the next period so it's it's a one period model but it you know it's not for ever okay uh similarly here decentralized the agents themselves uh set their own price after they've observed their cost so you have different prices here that the customers observe and they make decisions about um what to do and then the web 3 Model it's the same as the decentralized model but you remove the platform from the middle um essentially there's there's no commission and you replace it by a smart contract and you change the objective function that you're optimizing over it's no longer to maximize the profits of the platform because there is no

platform so why is this difficult to model because we don't know how people behave how they make decisions you know it's the it's the usual stuff right so you have to start from somewhere um and in the interest of time you know we start from this Workhorse uh linear quadratic utility model for how customers make decisions about consumption quantities etc etc so it's very well known in economics but it's for oligopolists um there's differentiation through this gamma Matrix here um but it doesn't really help us very much for the platform setting for one in platforms you have a lot more people so you don't need to worry about discret individual players necessarily um so it's not a good model also there's follow-up work

that shows that it's really hard to work with uh from Lina and cirar in 2011 Etc ET so we did a bunch of stuff and I'm going to wave my hands a little bit here even though this is potentially from a technical perspective quite interesting we do a series of of approximations one of them is innocuous it's just moving to continuous space so um sellers now are described by continuous distribution of cost as opposed to discrete then we have some meanfield types of approximations here you do all this and at the end you end up with a demand model OKAY like Coro you know berron the these types of things but it looks a little bit different than the usual ones and I should probably spend a little bit of time talking about this

one because it's going to drive you know all the results I'll show you depends on one parameter in this model okay so what does the demand model look like I think it's intuitive but let me know if it's not so this is the demand q that a single seller agent uh receives okay on the platform as a function of the price that they decide P which is a function of their cost and is a function of some sort of aggregate metric across the platform that's kind of a standard type of of thing now the aggregate metric I'm going to use P bar this is the average price across the platform and so when you write this out you end up with a with a intercept minus a linear term which should be familiar which just says

that you know the higher the average price across the platform the less customers you get okay and it's modulated by Beta it's just a sensitivity parameter so this goes in the right direction for the most part um what's interesting is this new term which is a little bit new compared to most of these demand models that you find in the literature and this model captures competition between the different agents on the platform okay so how does it do that well it uh looks at what the average price is across the platform and then compares the price set by this individual seller so if the price is lower you end up with less demand than another server who would have a you know sorry more demand than another server if

the price is higher you end up with less demand um so customers are basically comparing you know they have an idea of what the average is in expectation and they're looking looking at individual servers how they're pricing their own things their own products and depending on how far away you are you get either more or less demand and the important parameter that modulates that is gamma okay so gamma is how sensitive customers are to price differences in the market sometimes for instance you take an Uber you're going to pay whatever you don't care sometimes you take an Uber you're very sensitive to the price you can wait sometimes you're impatient sometimes you're not right so this is what this parameter gamma um is capturing and it maps to these uh

products that are substitutes that's the competition case versus complements I'll show you that in the next slide one last thing every paper uses aggregate metrics that are usually total quantities served or average prices median prices things like that we changed it in the paper we use weighted average price weighted average price means that you weigh by the quantities that are actually served and this is a point of content with the referees right now maybe you can tell me what you think I'm trying to explain to them that this is more intuitive for the following reason imagine you're buying a house you need to figure out where the market is do you care more about the listed prices or the transacted prices

that just happened I'm genuinely asking because I okay transacted is everyone on board with that or no you care about both potentially right but the transaction prices have more information in them because a trade happened so when you weigh by quantities you can actually take that into account if you don't then you lose that information you end up with all sorts of weird stuff platforms could use pricing as a gaming mechanism to post fake prices that they're not planning to serve and things like that it opens up all sorts of weird behavior that we don't see in

practice okay so this is the whole model um the integrals here are just sums because the platform doesn't observe uh sorry it's across the sums of all the costs across the whole uh Market but there's a cap potentially which is that not necessarily every seller is going to want to enter this Market it's a free entry Market but it might not be worth for everyone so we we needed a parameter take into account the fact that you know beyond some cost people will not want to enter so that parameter is called SE hat so everything's going to be a I'm going to fix beta for the rest of the um you know the talk and everything's going to be a function of gamma is it clear what gamma is or or not should we talk a little bit more about it maybe no so high gamma means a

lot of competition people are very sensitive to price differences low gamma means they don't care about price differences okay um one thing you might notice mathematically it's kind of nice but difficult at the same time is that the Q depends on P bar the P bar depends on the Queue so you have this feedback loop uh you have to be a little bit careful to make sure that that the problem is actually well well post okay it's not completely trivial this is just an explanation of how high gamma M maps to substitute products versus you know uh low gamma maps to complement products we could skip this as I'm a little bit behind qu question yes so I mean you're drawing like networks on some of the

like slides like this but like in the model the there's no network structure actually in the model itself right no we don't really need a network I mean we don't study the network topology um it's not part of the it's just everybody affects everybody B yeah it's not there's no local you know like local market there's no geographical um issues and things like those are all very important and interesting in practice but this is like a first pass type of model at this question so yeah it's a good point um is there any reason to think I mean my instinct is that the network structure might like buyas even more in the direction you're imagining is there any reason to think it might detract from the qualitative conclusion which direction am I thinking about though oh sorry like more more more towards the

the separation you find between when centralized works and when web three works right um I don't know actually I haven't I haven't thought about it um I don't I have an idea but I don't want it might be wrong so I prefer to keep it to myself here's some math how do you solve this problem so in the centralized case it's it's actually it looks quite easy so the agent you know individual seller has a a profit function that's quite simple it's their price minus their cost that's their margin but there's a part of that that's going to the platform that's what the FI uh the FI term is there so that's going to the platform so that's the per unit profit and then you you uh multiply by the uh equilibrium demand that they receive which was given you know by this

function in the previous slide which depends on P bar which depends on the prices that everyone else is setting you know there's no network structure but there's definitely you're definitely taking into account what other people are doing um and the break even free entry condition is here so if this is positive it defines the marginal um seller with a cost above which you know uh doesn't make sense to enter the platform's problem is here they they set a single Price p for everyone right so the platform controls here in the centralized setting and um they have they get some commission per unit and then you just multiply by the total number of units which is the sum across all the cues um which are parameterized by all the C's that are in the market the costs that are in the market so

that's the centralized case the decentralized case is similar with one big difference uh the there's no free entry break Break Even condition here you replace it by the fact that each individual seller is solving an optimization problem over their own price function PFC right so they pick the best price that maximizes their individual profits and then the platform's profit can be written similarly to before so now it cares about the average price across the market Market uh times the commission cuz it's the same commission for everyone that's another simplification and then you sum across all the actual quantities but the quantities now depend on the solution to the previous problem so this like a stackle BG type of situation where you first solve one thing and then you need to to use that

to solve the other thing right so are these even well defined yes go ahead there's the agents aren't heterogeneous so in equilibrium the average price like that thing they are heterogeneous where's theog it's the gamma it's this thing here because it dictates to what extent customers care about price differences plus the fact that their costs are different I see okay so those things are all indexed by some okay it's not an in when you go to continuous space you don't need to index it you can just name it C and it's a draw from a distribution is one Gamma or a distribution of a gamma the gamma is fixed which it in practice probably it's

not so pricing heterogeneity is coming from these C's yeah the heterogeneity from the sellers the sellers themselves have different costs have different costs we know their what distribution those costs have drawn from yeah that's what's creating some of the interesting pricing Behavior here if they all have the same cost I think the whole thing kind of um collapses prob I mean I don't know it seems like the model would be virtually the same if they all had the same cost you're going to have to choose a price and yeah yeah bangang instead of smooth they would all choose the same price yeah yeah exactly uh so there would be no price dispersion and price dispersion is one of the reasons that we get bad equilibria but I'll I'll show but it's possible I another thing I don't know

right so it's it's I I would have to investigate that we should probably look into it um as we're revising the paper right now it's a good point but like qualitatively there's a question of like do the high SE do the guys with high seas participate at all right would require some non gamma I would think right it's kind ofly yeah and so there are there are conditions under which you want them to participate actually but there are many conditions which you don't depending on gamma so um just to to try and and move a little bit forward um there is a solution to all of these problems and if you assume the distribution of costs across all these different sellers is actually just a uniform distribution you end up with closed form Solutions if you don't do that you don't have closed form solutions to these problems um and

they're actually really really simple and neat I know the font is too small you can't see it anyway so let me just show you visually what the solution looks like so it's it's Unique or it's Unique okay yeah it's Unique there's always going to be three plots again in the interest of time let's just focus on platform profits which is the first one always on the left so here I'm plotting uh profits as a function of this competition parameter gamma you know High competition low competition but I'm normalizing with respect to the optimal outcome under centralized pricing so I'm going to call the profits under that equal to one and then everything else is going to be a a fraction of that um and so the first thing you'll notice is that

centralized pricing does not depend on gamma this competition parameter why because there's just one price that everyone gets so they cannot compete on price in a given Market at a given moment in time that you know the sellers are not competing on on price okay now the blue thing is decentralized priceing now you give control away now everyone prices their own rides they solve their optimal problem and you end up with this blue curve where actually on the edges the whole Market collapses okay so why does it collapse is because they can't coordinate they're solving individual maximization problems it's you know a pretty classical issue in economics you end up with

coordination failures there's a Goldilocks region here which is a little bit counterintuitive and potentially interesting it's the fact that by giving up control of pricing the platform's better off in this small relatively small Zone okay why because there's no information asymmetry in that zone that's taken care of by giving up control so you don't incur the information asymmetry frictions at the same time there's no coordination failures in that not many failures in that zone even though they're self-interested it turns out it works out fine it's not a big problem um okay so the the magnitudes are kind of sign ific it's like 12% at

the best possible case of decentralization um not they not trivial but they do depend on some some of the parameters that we use Etc what's the takeaway so for me the takeaway was okay it's it's nice that decentralization can actually can actually help in some cases but the problem is that competition is not a static thing it's extremely Dynamic right so this parameter gamma I think Tim mentioned is there a distribution what what's this parameter it could change every day potentially what you can't change is you cannot go from uh giving control to sellers to taking it back impossible agreed it's virtually impossible to switch I'm getting zero

nods right I mean the burden you'd have to place on the seller is like one day is like okay set your price the next day no no I'll set it I yeah okay so what can we do uh we can try some mechanism design a lot of people uh know if you know if there are any people online or in the room that have an done mechanism design just a quick a quick word on this so we forget about the commission contract we say is there a better you know what's the optimal contract you search the Infinite Space of contracts but there's one thing that can help quite a bit which is the Revelation principle essentially says it suffices to ask people to tell you their costs just ask them and then you make sure that what you offer in return does not lead you

know give them incentives to lie about their cost in a nutshell okay so we're going to look at a mechanism where you ask people what their costs are and then depending on what they say you will say okay then this is the price you will set your product at this it's a centralized mechanism okay this is how much customers you're going to serve this is how much fee you're going to pay me you choose everything the platform chooses everything um but the first step is tell me your cost and so the way you write this up is you take into account the possibility that the cost reported to the platform from any individual agent might be a lie they might you know say it's C Tilda whereas in reality it's c not C Tilda so uh you write this up you

know the same types of stuff that we have before it's a little bit different because now the platform is has this generic fee structure WC that I call it here and it's trying to maximize the sum across the fees across all the possible agents in the market and then these two constraints the first one says that if if an agent lies and deviates away from their true cost they're worse off from an economic perspective they make less money but less profits and the second one says that they should still participate and so this would be a pretty standard mechanism design problem the one you could open up one of those textbooks and kind of follow step by step how to solve it that's what we initially did the one caveat is that there's this endogeneity issue here there's this P bar that depends on a

bunch of stuff so it took us a while to actually figure this out um but there is a solution for this as well and it's also unique and it looks like this it turns out the optimal structure is not a commission it's a quadratic uh function of your cost right so you Reveal Your cost to me and I will I will tell you your fee is going to be a quadratic function with a bunch of parameters that depend on beta and gamma in the market so how much competition there is in the market yeah so everyone should use quadratic you that's kind the conclusion which is not a good conclusion but let me at least show you what you get um so right now the blue and the gray are the

same as before I just zoomed in a little bit so it's centralized and decentralized the green is this new contract that we identified through mechanism design so what does it do it removes all the coordination problems and it gets you the maximum possible profit you could get for the platform so if you do it um you'll get you know more profits you'll get more Supply you get more customers served okay but you can't really do I mean I don't know if you know anyone that's using quadratic contracts for fees I'd love to talk to you at at the end of the the seminar so no one's using this so I just want so the the in this contract as the cost goes up the fee would get really aggressive so like someone would double the cost charging

like I'm not entirely that's a good point Tim I'm not entirely sure because these parameters could turn negative so it's not obvious to me that it's always that that case is um it's it's nonlinear so it you know it could could be weird so here's how we we went around that very quickly I know I see I'm running out of time so what do we do so you can't use cost I don't buy the story of people revealing their cost personally instead what we looked at are what are the other observables that people make if you decentralize and you can observe the prices they're setting and the quantities they're serving and so the idea is okay forget about asking for their cost let's observe what they do

and we're going to create contracts structured around the actions that they can take and this is what we call these apine contracts when um they take an action XI that's observable to the platform the fee is set as a function of that action and it turns out it's an Aline linear you know Aline means there's this extra term here so what actions can we observe one here's an example we can observe quantities and so the right contract in this context is forget about commission let's observe how much you're actually serving how many customers you're serving and then the fee you're going to pay the platform depends on that it's going to be you know the amount of customers you serve times something which depends on

the level of competition in the market plus a constant so this is what we call aine contracts for lack of a better word it's a little different so there are we looked at the literature there's something called two-part tariff program uh contracts these are a little bit different because this is on a per unit basis here so on a per unit serve you you you charge something okay so that is potentially more implementable I looked around no one's doing exactly this in practice but there's there are versions of this that you might have heard of things like hey if you if you you know double your sales I'll give you a discount that's what this is saying right kind of okay so how does this thing actually work right this I thing

goes back Tim's question how is it achieving these gains over these suboptimal commission contracts Etc so what it's doing is it's it's affecting the incentives that each individual agent has to join or not join and serve customers and so these are two different Market situations let's just do the left one these are the total uh fees paid by each agent as a function of their realized cost which means that let's say I'm an agent my realized cost is 04 I'm here this is the amount of fee I'm going to pay the platform however there are negative fees which means there's subsidies so the optimal contract does wealth transfers between different types of Agents some of them you give them

subsidies to encourage them to serve more customers other and and you collect fees from others and it can switch depending on what type of Market you're in if you're in high competition versus low competition this thing can completely switch so it's not at all trivial what it's doing it looks kind of simple with the offine structure but when you actually look at how it's doing and achieving all these gains you know it's it's not trivial you have all these these subsidies and um and transfers okay now um I I know I'm going a little bit quickly here but this is what you all came for probably what are the implications for web 3 I can't fully solve the web 3 problem because I you know I haven't done the tokenomics yet but what we can do are

some first order things that are important such as let's explore what Jens is doing in all these other startups let's just set the commission fee to zero and see what happens but we can also solve for the optimal contract in this context without the platform by changing the objective to instead of maximizing the platform's profits we're going to maximize the collective amount of wealth that the agents are making so that's this new objective function you know let's skip through it um it's it looks a little bit different than before and the result is that it's the same structure that's optimal so this offine structure remains even in this setting the parameters are completely different so the way you do wealth transfers might be very different

between the different agents but the structure is the same and this is what it looks like visually the green now is the optimal platform contract that you could either Implement you know through the centralized quadratic cost contract or you can Implement to decentralize with by observing the actions that they take like the quantities and things like that regardless of how you implement it these are your optimal profits as the platform and if you're let's say a startup like Jensen and you say you know what I'm going to compete against these platforms even when they're optimal I'm going to use commission contracts sorry I'm going to use um um zero commission contracts without creating a new contract so I don't take any money away you end up

with the same problem as before where because people are self-interested they're not coordinating with each other so you're not resolving the initial problem was there to begin with in decentralized pricing okay so if you're just going to do zero commissions and let people figure it out set their own prices it's not going to work for all markets and the problem with startups this is me taking a little bit of a leap of faith is that you're not really in the Goldilocks competition Zone if you're a startup you're typically in the extremes like there's not enough competition usually you know um or you don't get enough sellers to compete with each other that that's what I mean okay now if you implement the best one right

the optimal contract with a you know without the platform you get the Holy Grail of web 3 all the promises that it's been doing from the beginning you know you remove the intermediary so there's more value to be spread out across the users of the system you get a bunch of gains from that and the fact that there are no more frictions you get more gains from that too and so you get this massive boost um if you could do everything optimally 33% more profits you you double the amount of people that are serving the market uh you double the amount of customers you know so it's all the promises that we've we've heard in this I you know idealized setting basically so what's the difference between the optimal web 3 and the basic web 3 yeah so basic web 3 is just

removing the platform but not solving an optimal contract so the optimal contract does Fee redistribution so imagine a smart contract that for every P every seller say okay you set your price but I'm going to charge you fee based on how many people you Ser Quant how many customers you have that fee goes to the smart contract get redistributed to another agent okay the the there should be no you know balance in the smart contract in terms of fees it always gets redistributed out what does welfare look like for the sellers under the optimal contract versus the so this is it because there's no there's no um there

there's no platform in the web 3 setting but this there must heterogen across the sellers in welfare from this contract right yes so um I think there might be situations where not every seller is better off compared to some other Benchmark but as a group they are it's like an interesting question about decentralized governance about whether or not you could implement this contract bye yes I I think I have some slides in the appendix but I'm not going to have a lot of time to go over them so Ju Just to to wrap things up so there was a question about tokenomics so what this paper is missing is I've shown you a potentially optimal contract you can buy it or not buy it you know we could do some robustness checks and all that all that kind of stuff the question is how do you actually implement it

what's the role of tokenomics in implementing that in practice I I I don't know the answer to that that could be an interesting thing to do we have an older paper here that looked at this uh I think the question that was raised at the beginning which was you know how do these people make money right how does Jensen make money and other startups like this um and it's a question of token retention versus commission contracts which are you know used by non web3 platforms and so we find in this paper it's a different model though we find that commission contracts are actually better from a profit maximization perspective compared to token retention um token retention is worse but it actually leads to better service levels means you get more

Sellers and more users on your platform even though you're giving up some profits as a Founder commission does not dominate universally because there are situations where token retention can actually help you know under moral hazard and so there are situations where you might want to use both in the paper it doesn't happen very often but it does happen and the limitation of this paper is that it doesn't take a mechanism design approach so ideally what would be nice to do is to merge you know some of the lessons learned from this mechanism design paper together with you know some of the things in this paper that deals with tokenomics and try to actually bring these things together and you know potentially uh help startups with some uh recommendations okay so uh just to recap does the pricing structure it's the same slide as before right so does

the pricing structure matter the very least I hope I've convinced you that the structure kind of matters and there's this core you know economic tradeoff between information frictions coordination frictions there might be other things you know but in this paper there's just this okay um neither one is optimal you can do better by exploring new contracts something that we call apine contract seems to work quite well um and in terms of web 3 uh you need to be careful how you implement it the way most startups are doing it right now you know it's you you you're exposed to just market failures basically because if you don't coordinate people

even though you get this extra boost because you don't have commission to pay to anyone it still doesn't work right um on the edges so you have to be a little bit careful but there's a lot of promise there [Applause]