The Value of Decentralization Using the Blockchain with Marco Reuter | a16z crypto research talks

Transcript

we are thrilled today for our seminar to have um Marco reuter um from Manheim he's a graduate student and uh you know I you know interacted with him in his work for a while through various different Market design channels he's got quite a broad portfolio across uh different Market design projects but in particular when he when I first saw what is his job market paper footnote he's on the job market this year uh so you know can't say that often enough uh when I first came across this job market paper it was clearly something that we had to see it's um you know really trying to understand a foundational question and uh in in why we use crypto and and we're thrilled to have him tell us about it welcome Marco yeah thank you so much

thank you so much for having me Scott um I I tried to share my slides okay yeah you can see the title of the paper is the value of decentralization using the blockchain and as Scott had said it's my job market paper so any feedback is uh extra welcome so let me just uh start right away um so the question that I want to ask is you know suppose you're having a situation where you're thinking oh maybe I should make the new Facebook or the new Google right I think at the a16z they oftentimes encounter projects maybe that claim to do exactly that so the question that I'm going to look at in this paper is should you do so using a regular company just you know make a startup and have a regular company

or should you try to do it in a decentralized manner through the blockchain or through a blockchain and what are the key points of the paper the key friction is that I'm going to assume that users suffer from a locked in effect so I join a platform maybe a Facebook or Google and you know when I want to leave again maybe I encounter some switching costs so that's the cost that I have or you can think that I you know I search on Google every day and Google gets kind of to know what I like to search for and then you know I find that nice and then it's harder for me to leave so I'm I become logged in over time and the key economic problem that is

going to arise is that the entrepreneur who makes the platform is going to be tempted to exploit this locked in effect for example by putting my advertisement so you know I make a search engine and I start off and I have only the first result uh it's advertisement and then over time I put the second result and the search result and you know I think you know intuitively kind of what's going on here or I try to sell the data and I lack commitment that I won't do it so I can tell you today no I'll never do that but you know later down the line if it becomes profitable I will do it anyways I can't really commit that I won't do it and the main result of the paper is that I can use blockchain to generate such

commitment and that as an entrepreneur who thinks about making such a platform I should decentralize through a blockchain if and only if this locked in effect that the users have is sufficiently large so are you assuming there are network effects to using this platform for for users yes they're they're going to be Network effects in the model um yeah if you have any questions at any time super happy to just jump in so let me go to the next slide just listing a couple of applications I think you know usually I have to explain a lot more about blockchain but for this crowd I think um it's good to just jump into the application so in recent times many

decentralized decentralized alternatives to traditional companies have popped up so on this list on the right hand side I've put the normal companies that I think you all are very familiar with like AWS Google Facebook Twitter Spotify and so on and on the left hand side kind of some decentralized Alternatives that have been popping up like a file coin or pre-search or you know these kind of things and then off now the paper goes into answering the question of okay why do we observe that why why should I do it on uh on the blockchain and kind of the key feature is that for this decentralized Alternatives I'm always thinking about decentralized governance which means that I'm trying to put the control over the network with

the users yeah so usually I'm going to have some sort of governance mechanism that is some sort of voting so if you have a project you make a cryptocurrency and everybody who has tokens can vote on what's going to happen I think lots of these projects have both quite regularly and you know what happens effectively is that the entrepreneur that founds the network gives away control over the network and that's going to be very important to you all right so briefly talking a bit about the literature so there's kind of two strains that are important here there's one economics of blockchains it's kind of a very recent literature and you know where I fit in is about kind of using

blockchain for commitment but what makes kind of my paper Specialists taking like this Dynamic approach of a grow my network over time and then I have these locked in effects and then there's a bit more of a traditional literature in the i o sector about platforms because the the networks or platforms that I talk about are platforms as such yeah but usually those are more concerned with pricing and platform competition but let me not uh get into the literature too deep so what I want to do now is kind of give you a high level overview over the model just to give you the intuitions of what's going on and then after that we'll go kind of dig into the details

yeah but this for now is a bit vague on purpose just to give you the intuitions of what's going on so what's gonna what's gonna happen the entrepreneur is gonna create a network and can decide do I do it in a centralized way through like a normal company or do I decentralize then it's a dynamic model where over time potential new users they keep arriving and the old existing users they're going to be logged in so they find it harder to leave and it's going to be profitable to exploit this locked in effect I.E I put more monetization in my network and what what do I mean with centralized is the entrepreneur just decides

monetization so I make you know a company like Facebook and the CEO wakes up in the morning and things putting more ads is going to be profitable we'll make more money that way and then he just does it that's the end of the question and in the decentralized world here in this model I'm going to say no the entrepreneur doesn't have power over doing that I'll let the users decide to do that so I'll have some sort of decentralized governance maybe using a smart contract where it's like the users then decide how much monetization should the network have so what happens in a centralized Network here I drop draw a picture because I think it's going to be quite intuitive so on the x-axis you see the time yeah

so it's gonna have multiple periods or infinitely many to be exact and on the y-axis you see how many users are in the network and in the equilibrium that we're looking at it's going to divide into two phases yeah there's going to be a growth phase and then there's going to be an exploitation phase so what happens in the growth phase is me as the entrepreneur I can't commit that I won't exploit you looked in effect but at the start there's lots of growth going on so you know I implicitly gained commitment that I want to use your locked in effect like it's just much better for me to grow the network but then as growth slows down eventually the temptation of just monetizing my

existing user base becomes very large and in the end it's going to become so large that I'm just going to say no I don't really need to get more new users I'll just monetize what I have and then I get into this exploitation phase so so Marco just so I understand the plot so in principle this blue curve keeps going and grow growing slowly yeah so this this is this is actually a decision by the entrepreneur to switch from the blue to the red is that right exactly okay all right so the advantages for the entrepreneur in the centralized network is well I have full control over the monetization but in some sense that's

also the disadvantage right because I lack commitment over my future monetization decisions like if I wake up tomorrow and I'll have two billion users in a network and I find it very profitable to monetize them I'll just do it and just there's no way that I'll be able to stop myself from doing that tomorrow then let's switch over to the decentralized network and I think connects the terms questions so now you see the same graph but the blue curve just keeps growing and so on the x-axis again you have time on the y-axis you have the users and the decentralized network is going to have the following Advantage I can commit that I'm not going to abuse the locked

in effect because I'll let the users decide on monetization yeah and with that I can keep growing over time but the disadvantage is that I lose control over my monetization decision and actually in the model I'm going to have to engage in Revenue sharing so I'll have to give away some of the revenue from the monetization from the advertisers to my users why well because they decide the monetization and if I give them nothing they will just say no no monetization please we'll just choose zero yeah so I have to give away some revenues to align preferences and you know lots of these projects so for example when you go on pre-search

and search like every time you search you actually get a bit of money so there's kind of Revenue sharing going on and also many other of these projects I have a question here um are you distinguishing like users who are just using the network versus like users you may like hold some tokens in building shares who have like nominal um like interest probably against another video or there's like the same party in the module in this model all users are going to be the same so there's not going to be any difference okay all right and although actually quick follow-up on that yeah it seems like your model could you could imagine every person we haven't seen the model yet but of using my slight amount of knowledge of what's in the paper already like it

seems like you could imagine a universe in which only some of the users hold governance um you know and and therefore they have some incentive to exploit the other users but they're subject to the same you know fee structure as as everybody else and so you might actually imagine that you can even get the the version of this result where there's like some exploitation but not as much as there would be under fully centralized control with partial decentralization is that the right intuition yeah yeah that's exactly correct right because the users they're still going to suffer from the monetization even if they're holding some of the tokens right so they will internalize some of the exploitation they would be tempted to exert on the others but uh yeah so you you could break this up slightly and qualitatively

nothing much changes awesome thanks all right perfect thanks for the question right so let me give you the main result of the paper just because I you know I really want everybody to take away this result and then afterwards we go into the details and actually see how you get there so the main result of paper is going to be the following preposition they exists a well-defined size of the logged in effect it's going to be u double star here such that decentralized governance is preferred by the entrepreneur I.E makes more money if and only if the size of the locked in effect is larger than this threshold and so if the logged in effect is very large decentralized if the logged in effect is

small stay centralized that's going to be the main result all right and now we can jump into the model and you'll kind of see how everything works and if at any point in time there are questions please just interrupt and ask all right so what does the model have there's one entrepreneur that's going to be the person who makes the network and there's going to be a Continuum of users I and in every period the amount of users in the network of course might change so it's denoted by mu t f in Period T this mu users it's a sequential game with infinitely many periods so I started period 0 1 2 and keep going on forever

and the periods are slightly different yeah because period zero is special that's the period where the network gets created and then periods one two and so on until Infinity they're all the same and what happens is that the users interact in the network and the interaction is subject to a couple of key things first of all the level of monetization so you can think of how many ads should I show in the network and it's going to be a real number maybe two ads maybe three ads and from this amount of advertisement that I display I can make revenues I also what are the revenues I have this monetization level pi and I multiply it with you know maybe you just think about

an advertising right that is going to be fine NFI is going to be increasing so if I want to advertise to more users I have to pay more I think it's you know quite intuitive and if I don't advertise to anybody I don't pay anything Revenue can be shared regardless of the mode of governance and the entrepreneur keeps the share Alpha T and the rest goes to the users and the Assumption here is going to be it goes to the users evenly so they're going to share evenly so so sorry Muse a measure Museum number I was a little confused um so it's uh it's going to be a number but it's infinitely many users right but and

then I assign a measure to how many users they are so it's not like a natural number of 180 000 users but it's a measure okay so for fixed T it's a number yeah okay yeah okay all right let me talk a bit about the timing and the actions so how is monetization so when you say fee is increasing you mean for each fixed T so it's increasing in mu sub T right I should think yeah exactly okay all right more users means I have to pay more for displaying advertising yeah all right thanks for the questions okay so timing and actions that's where I was right so I'm going to look at

house monetization and revenue sharing decided so in period 0 that one was special where the network it's founded the entrepreneur picks a mode of governance so I do it in a centralized fashion or I decentralized and then if I pick centralized in the following periods the entrepreneur just does whatever she wants so she just picks monetization level Pi so how many ads to display and revenue sharing Alpha how much of my Revenue do I want to share with the users if I do it in a decentralized manner I'm going to do the following in Period T equals to zero I pick Revenue share that I'm going to share with my users right here that's the kind of the tokenomics so the idea behind that is in

period 0 I make some sort of token and I keep Alpha of the tokens for myself and I give one minus Alpha of the tokens um to the community and then kind of the idea is I ask advertisers for example to buy tokens on the open market to pay for the advertisement so that we kind of uh evenly share in the price appreciation of this token it's a bit uh in the corporate finance world doing would maybe say it's like a stock repurchase in a sense and um yeah in the following periods then the users collectively choose the amount of monetization that they want so the pi and uh yeah that's how we choose monetization and revenue sharing

then let's talk a little bit about user actions and user growth how does that work so regardless of its if it's centralized or decentralized in Arrow the periods where the people use the network the existing users that can stay or leave so they have a binary decision and new users and they can say well I'll join or I'll not join yeah also binary decision and network growth is modeled as follows so I have a growth function so what do I do I take the graph function G and I look at how many users did I have last period and that's the amount of users that I start with at the beginning of this period and then the amount of users that arrives yeah the new potential users is

G of mu T minus 1 minus mu T minus one and that's some sort of positive number yeah so that's people that haven't used the network yet and they kind of just arrived to the network and if they want to join they will join so that the math works out the following way if everybody who arrives joins now the network size is G of mu T minus one and if nobody of those people join the network size stays at mu T minus again if there's any questions somewhere um please ask which you can mu of T decrease if people are leaving the network here yes yes multi can decrease if people leave so um

the existing users right they can always leave the network so it can decrease um what's going to be in this model is users are going to be homogeneous so if one wants to leave everybody will leave right so if they all kind of have the same preferences in this model it's either they all stay or they all leave um so in that sense it's a bit uh extreme here all right let's talk a bit about preferences so what does the entrepreneur want the entrepreneur is purely interested in Revenue right I make a large Network there's lots of money potentially to be made and that's all I'm interested in so what's the utility per period right they look at how much revenue do I keep

that's Alpha T and then I look at the revenues that I make and the revenues that I make again was piety the monetization how many adds to a show times the advertisement rate this Phi of mu t for users it's a bit more complicated and so there's going to be three parts to their preferences first is usage of the network so all right I use the network and that actually gives me some utility so if the network has measure muity of users then the utility for each user is V of mu t and V of mu 2 is increasing so I have some sort of network effects and if I'm the only person in the network so if

there's a zero measure of users I I don't find it appealing at all right if you're on Facebook alone that's not not really very appealing second part is monetization so nobody likes to look at advertisement I think that's quite intuitive so what happens is if I display Pi T many ads the this utility that users get is minus K times pi t squared so some sort of convex function here and K measures however semit monetization and so if K is very close to zero then I don't really care if there's monetization or if there's advertisement and if K is very large then I really hate it and I really don't want to have a lot of that

last part is yeah Revenue so if there's some sort of Revenue sharing going on I like that how much revenue do I get well users get one minus Alpha of the revenues right but we divide that evenly over the users so we divide by mu T and then I multiply by the amount of revenues that are generated in the network so that the utility of a user in the network is add some of the usage utility minus to this utility for monetization plus utility from any revenues that I can get okay and here comes kind of really crucial part is when I look at the outside option yeah that's going to

change now before I join the network my outset option is normalized to zero after I joined the network yeah I've have an outside option that is worth minus U smaller than zero yeah so there's again two ways to interpret this either I have switching costs and and they are u-large or I use the network for period the algorithm learns what do I like and now I get kind of this extra utility u from the algorithm Having learned so comparatively the outside option has gotten worse by you so that's the locked in effect that I'm talking about and agents so the entrepreneur and the users the maximize in pretty standard

way the sum of discounted utilities and there's a discount Factor Delta by which I discount future utility maybe that's a good point for uh stopping for questions in case anybody has any um does does this factor in potential competition from from other platforms that are providing like you know for the sake of argument identical services but you know with with less quantization and and therefore less disutility yeah so um the other platforms that implicitly

in here a bit for the outside option yeah but of course that's very implicit and very kind of reduced form so I'm not not really modeling explicitly what monetization do the other platforms um pick and so on like that would probably be a very interesting Next Step but it's not in this model yet so I can't comment on that much okay thanks all right thanks for the question um let's move on so what's the solution concept that we're going to use it's going to be sub game perfect Nash equilibrium and then we need to talk about strategies right so what's the strategy for the entrepreneur so I need to look

at the possible histories of the game so I need to look at what has happened yeah for the entrepreneur what's really going to be The Driver of the strategy is I need to look at how many users do I have yeah and these histories I need to map into the following yeah if it's a history that terminates at period zero I need to make a governance decision I make it centralized or decentralized and if it's a history that terminates in a kind of a centralized game I need to decide for a degree of monetization for the period and for a degree of Revenue sharing and if it terminates in a decentralized game well then I only at the start make my decision of Revenue

sharing and in the future users going to vote on monetization so I don't need to do anything and if the point here why sub gain Perfection is really nice is that what strategies you need to be sequentially optimal so I don't have any commitment of future monetization or Revenue sharing right I need to be sequentially optimal for every history so again if I wake up and for some reason I have 2 billion users and it's very profitable to exploit them I will do it because it's sequentially optimal all right then let's go to the users so what do I need to do for them I need to map the possible histories of the game and through a decision to stay or

leave if I'm an existing user in uh in the network I need to make a decision to join or not join if somehow at that history I'm just newly arriving and if the platform has decided to be decentralized I need to vote for a degree of monetization piping okay so let me give you a quick roadmap of what we're going to do first I take you through the centralized governance sub game so what happens if the entrepreneur decides to do it in a centralized manner then I'll do the decentralized governance sub game I.E suppose the entrepreneur decentralizes what happens and then once we know what happens in

both the sub games right we can look at okay T equals to zero what should I do as a governance decision should I stay centralized or should I decentralize right so let me give you the exact timing of a particular period in the centralized governance sub game so if the entrepreneur chooses centralized governance every period has the following timing first the entrepreneur chooses the level of monetization and a fraction of Revenue sharing for that period users can observe that for that one period and then they make it simultaneous Choice everybody who has arrived the net period decides to adjoin or do I not join and everybody is

already in the network decides to a stay or do a leave and then utilities realize for that period all right before we can kind of say what the equilibrium is we have to think about a couple of preliminaries and one preliminary that's going to be quite important is the following so look at the entrepreneur's problem to maximize Revenue just myopically for one single period where I want to make sure that every user has a utility level you had from using the network while the network size is muted as for that consideration is going to be quite

important and you can calculate it but the details aren't super relevant here so I'll just say We'll denote that um that Revenue that I can make with mute users that are supposed to get you had utility as PSI of mu mu T and you had and so that's definitely from them sure I think if you had as like the threshold where the existing users are indifferent between staying or leaving or um so for now it can be arbitrary but of course the thresholds are going to be the relevant utility levels that are set as the entrepreneur right so can be anything in this kind of formulation here but yeah you're very

very right it's going to be the thresholds that will be relevant okay um then I make one assumption in the model which is the following so I take Phi T Phi of mu T divided by mu T so what was that I take the advertisement rate and I divide by how many users I'm advertising to and that's not decreasing um what does it do in the model well it ensures that if I have more users actually want to grow because I can make more revenue from that and of course it's important to look at well when does it hold when doesn't that hold so if you would put a very standard cost per view or cost per click advertisement

model on there you would actually get a constant fraction uh there um and that's kind of fine for the Assumption uh so kind of for very popular models of advertisement this uh this works okay second preliminary that we should talk about is well how large should I grow my network right I'm uh I'm a centralized entrepreneur and I always have this temptation of abusing reluctant effect but I don't want to do so right away with zero users because then I don't make any money because you know can't sell advertisement so I definitely want to grow a little bit but the question is how large exactly for that we look at the following

definition that we're gonna and that I'll explain to you so what I'm going to Define here is a threshold Network size mu bar and this threshold Network size balances off the following two considerations the first consideration is I have this expression right so PSI was the function that gives me profits per period and this right coefficient of 1 divided by 1 minus Delta that comes from geometric Series right I do this for all eternity so I get that much in today's value and what do I get well I have mu bar users right because I'm not growing

anymore and I'm exploiting the locked in effect and then you know it goes into the direction of the question that timid asked well how much utility am I giving right and the answer is well I'll just keep them indifferent between staying or leaving right so if they leave they have to pay you in switching costs so if I am give them utility minus 1 minus Delta times 0 that just makes them indifferent between um staying and leaving so the other consideration that I can have is well maybe I'll grow the network one last time before I exploit this locked in effect right then I have two components here again the PSI functions the revenue

functions are going to be the important part you see the first argument of the function is not mu bar anymore but it's G of mu bar so I grow the network one last time yeah the second part of the addition here you see I do exactly the same I give everybody a utility level of minus one minus Delta U to keep them different between staying or leaving and but today when I grow one last time I need to compensate and I know that today when they are freshly arriving at the network the outside option is worth zero and if I want to join I need to give them at least your utility they know in the future they will be exploited right so they will get minus U so today I need to

compensate them for that so I give them Delta times you in utility and now for this talk um I will just say suppose mu bar is well defined right in the paper I go into much more detail with that and kind of the intuition that you should have in mind for when mu bar is well defined is that when growth slows down over time I will get to this point and so maybe at the start I have very rapid Pro growth so kind of growing is always going to be very nice so you know not going to have this problem that I would rather exploit but when growth slows down over time eventually I will get to this point uh mu bar where I

would uh you know in the future rather exploit them and not grow anymore so that's the intuition behind when is Mu bar well defined I forget Marco do some assumptions on G um not at the start so G is just kind of increasing people come along but for Mu bar to be well defined G needs to be decreasing over time and that's maybe not very satisfying because it's you know not a very mathematical assumption but the kind of math behind this is a bit more involved so I would just want to push the intuition that it needs to decrease over time all right

um let me move on to what's the equilibrium right now we've gotten some preliminaries out of the way now we can look at what is the equilibrium so following proposition suppose that some condition holds for now ignore it I will tell you afterwards why it's important and what happens when it fails then I get the following strategies for a subkin perfect Nash equilibrium the entrepreneur's strategy is going to be in three parts the first part is as follows and what decides the parts is you see here the amount of users that are in the network so if I start with an amount of users that is so small that I know that tomorrow I will still want to grow even though I don't have commitment

I just know that it's tomorrow still going to be optimal to grow I set monetization and revenue sharing to maximize Revenue given by the PSI function and I do it with the following two parameters I give a user utility level um you had of zero and I do it for Network size where I anticipated everybody is going to join yeah so that's part of the growth phase all right then the second part is now the network has grown so large in this part here that I know that I'm growing one last time now I know I can grow one more time and after that I will just find it much better to exploit so what do I need to

do I need to set um a monetization and revenue sharing such that everybody now gets a utility level data you right I'm compensating because they will be exploited in the future but still everybody is going to join because they're indifferent between joining and not and the tiebreaker here is they will just join if they're indifferent and the last part is now my network is very large it's larger than mu bar so I'm not gonna grow anymore so what do I do we've seen that also before in the trade-off I said the user utility level of -1 minus Delta times U and I do it for the network size where I know nobody's going to join anymore it's

not good for the users to join anymore so the entrepreneur strategy if you think back about the pictures that I've shown you at the start has these two parts to it there's one growth phase that are the first two bullet points here and then there's the exploitation phase that was the red part that's the last bullet point here now so just keep the picture basically in mind that that's the mouth of the picture then for the users the strategy goes as follows in the period when I arrive to the network I will look at how many users there are right because that's important to know whether or not the entrepreneur is going to exploit or not and if the amount of users are rather small I anticipate that

the entrepreneur is going to grow in the future so I'm fine with getting a userability level of zero yeah but then the second part of the strategy is oh I see the network is quite large already so the entrepreneur is going to exploit in the future so I'll only join if you compensate me for the exploitation of the future so I want Delta u in utility and if I'm already logged into the network well then I'll stay basically if I'm uh indifferent between staying or leaving and what does that mean I need to get a utility level of at least this minus 1 minus Delta times U so that if this always happens in the future right

I have minus U and have a difference between staying and leaving um so I think the equilibrium in math may be a bit difficult so if there's any questions please go ahead I mean I guess one of the things that's like wacky and kind of special about the the framework here is that you know outside of an equilibrium framework if you is sufficiently large then you want to exploit the users as much as possible right if you don't have to worry about like users being you know rational about uh joining the platform uh then you would just you know exploit them as much as possible the more and the more locked in they are the more you can exploit them um I guess if you have an effect where the network effect of the platform grows

essentially the outside option gets worse and worse the more people are on the platform um that compounds all the effects you're showing here right that somehow like people should be like in equilibrium people should be even more concerned about joining because they sort of you know inter it's not introspect exactly but they understand an expectation that it's going to be even harder to leave than they might think or something like that yeah that's a great point so right here the kind of outset option like audio locked in effect is fixed but I can make it dependent on the users and if it gets worse worse with more users then as you had rightly said actually the commitment problem gets worse so yeah basically you can do it but qualitatively the model stays the same it's actually the predictions

it's more than qualitatively the same it's actually like you're you're tying your hands against um you know against the finding cool thanks yeah well perfect thanks um okay last point that I want to make is right in the preposition you see at the very start kind of I have suppose that kind of these parameters are larger than Delta U right so Delta U was the amount of utility that I basically give to compensate before exploit the users and if this inequality doesn't hold I actually don't make enough Revenue in the network to pay this compensation and then what happens as an effect is that if this view is sufficiently large so that the

inequality doesn't hold actually nobody will join an equilibrium because they know I can't pay the compensation so for really large levels of you actually I don't get any users and that's really bad for centralized governance already there were centralized covenants um next up is decentralized governance so um how does that sub game work and let me again give you the timing to catch you up what happens so if the entrepreneur chooses decentralized governance in t equals to zero she's gonna pick this unchangeable level of Revenue sharing iei keep half the currency or keep 40 or whatever and I give the rest to the users and

actually that I do it at an unchangeable level is without loss of generality it's not really obvious that it is but in the paper I kind of shown an extension that it actually is so um for here just think doing Alpha constant is uh with our loss of generality then the following periods have the following timing users again they make a simultaneous Choice those that are not present in the network iados that freshly arrived they choose to join or not join and those that are already present in the network they can stay or leave then everybody's in the network now that wants to join or that doesn't want to leave and then users collectively choose the amount of monetization that they

want to have for the network and their utilities realize okay here we can go to the equilibrium a bit faster why well because we can focus on equilibria and undominated strategies so what does it intuitively mean well users in this model they're actually homogeneous so they should actually be able to agree on what do they want so if I do voting here I'm just gonna tell them vote truthfully or vote as if you were pivotal and there's a bit of a mathematical reason behind that as well namely if I vote in the voting game with a Continuum of users I'm never going to be pivotal so I don't know if you're not familiar with 14 games usually you have

to condition on being pivotal and then make an optimal decision but in a Continuum I'm never pivotal so I can basically do anything and then for this game I should just do what I actually want to happen and what is it that I want to happen is the following I find this up game perfect equilibrium such that one newly arriving users they will always join the network so I don't have these growth and exploitation phases where they can just keep growing and every period the users of the network will vote for a degree of monetization that is this Pi T star here yeah and what does it depend on of course it depends on the amount of Revenue sharing

that the entrepreneur does and so if the entrepreneur keeps 100 of the revenue for herself that means Alpha is equal to one and in this expression you plug in Alpha is equal to one users actually say no monetization please right I don't get anything from that and it's just annoying please no ads so the entrepreneur is actually going to have to share revenues to get a positive amount of monetization and if you do the math and the model here turns out that I want to share half of the revenue with the users or if you want to think in terms of like token right they give away half the token that I have uh to the users and the network and again this Revenue sharing is

necessary to align the preferences right if I don't share money and the users get to decide monetization they will not vote for any monetization so I share some here it's half two line preferences all right so now what we know is what happens in the centralized governance sub game we know what happens in decentralized governance sub game right so now we can go to the kind of big question that the paper had should I centralize or should I decentralize all right go back to period zero let's recap what we know so far in the centralized governance sub game I had argued or I shown you that when U

gets very large in this kind of condition and the equilibrium didn't hold actually I don't get any users right so network with no users that's going to be very bad so there's no Revenue so kind of that's the first indication of When U is very large I should decentralize why well decentralized governance at shown you always has some Revenue because it always has users and it has user growth so the first conclusion that we can draw is When U is very large I should decentralize foreign how does that generalize right we would you know want to know much more not just what happens for very large use so it's actually good to look at the other end right Suppose there is just no

logged in effect there's no problem at all right and if there's no locked in effect to exploit decentralized entrepreneur actually doesn't have any problem and you get kind of the following result it's you know not very surprising that as the size of the locked in effect goes to zero I always want to stay centralized why I don't have a commitment problem right the downside of centralization goes away but I keep control over the monetization and I don't engage necessarily in Revenue sharing so that's better for me for sure so yeah there's no reason to decentralize if there's no commitment problem so now what we know is kind of the two ends of the spectrum U equals to zeros

stays centralized U equals to large enough however large that may be decentralized but we don't know the in-between yet and the in between is also kind of interesting part so now we want to cover the in between all right this is kind of recapping what I said very large use decentralized very small use stay centralized what about intermediate levels and this is where we get back to the proposition that I've shown you at the kind of a medium part of the talk after it shown you the pictures is that I can find a well-defined size of the locked in effect you double star such that I should decentralize if and only if the size of the locked in effect is larger

than that and this is kind of the only part of the paper that I really also want to give you a proof sketch because you know that's the cool made result so I think we should find out a bit how does it actually work and for the proof it's going to be very useful that we know both extremes right we know at U equals to zero as a i centralize at U very large I decentralize and then what we do is we make the following observation actually the revenue and the decentralized network is independent of you why does that hold well users don't abuse their unlocked in effect right so for them it actually doesn't matter how large it is

so the revenue is actually independent of that and then the second step in the proof goes as follows I look at Revenue in the centralized Network and I show that it's actually decreasing a new and you know then I start above right at zero and then I decrease and I end below and if it's kind of continuous I get a I get a point of indifference I get this threshold so now why is it decreasing there's two effects that we should discuss there's a first order effect which goes as follows right in the last period of growth I had to give utility level data times you to the user so that they will join

and what did I get from exploiting in the future well I was able to extract -1 minus Delta times U every period in the future but kind of what it turns out is that actually the revenues that I make are going to be concave and how much I can extract from the users so that it's actually going to be more costly to compensate up front compared to what I get out in the end from exploiting so that's an overall negative effect for Revenue and kind of the second effect that I have is that a smaller Network size actually means that I have um that I have a smaller revenues and so um how does that work well so I had this

mu bar this is kind of the size of how large I grow my network and what I show in the paper is that this mu bar is actually also decreasing in um in the locked in effect so for a larger locked in effect the Temptation becomes becomes increasingly large so that I will um start exploiting earlier I will grow less tall or less large my network so mu bar goes smaller I had a smaller Network size and that's also bad for revenues So eventually here from that intuition I start above with centralization at U equals to zero then it decreases and it's going to be continuous here new and

eventually I'm below and that's how I get this nice cutoff all right that was the main result so um now I want to quickly talk about welfare right there's always something we're interested in as well so far I just talked about what was what does the entrepreneur want to do but um maybe we should also look at what should one do in terms of welfare here first I want to look at the Criterion of Pareto efficiency right that's usually the minimum that economists care about and if something is not part of efficient you know we get a bit angry and say oh we should be very efficient so I get the following

corollary with the analysis that we've done so far is that decentralized governance of the network is a Pareto improvement over centralized governance if and only if I have the same relation that the size of the locked in effect is larger than u double star and the reasoning behind that is fairly simple well the users they always like decentralized governance more right because in centralized governance they actually always indifferent between joining and not joining and so they get zero utility overall while in decentralized governance if you do the math a bit it's not hard to show that every period I get positive utilities so users always like that and then for

character efficiency I just need to look at well who else is in the model the entrepreneur and then I get the same threshold right because only the entrepreneur matters say for operator efficiency um to decide which mode of governance is better then the second welfare Criterion that we're usually interested in is utilitarian welfare I.E we just sum up the the utilities of the users and the entrepreneur and first result there is of course if we have a period to Improvement right we also get a utilitarian welfare Improvement so if the size of the locked in effect is larger than you double star we should also decentralize from a utilitarian point of view

but if the size of the locked in effect is smaller then actually the effects ambiguous well why is it ambiguous well users get higher utility I had argued for that but the entrepreneur gets lower utility because she makes less revenue and then it really depends on you actually have to run the math and you have to make some functional assumptions and kind of calculate which effect is stronger here ads in general it's kind of hard to say so the effect is ambiguous here can go both ways all right let me conclude what did I ask in uh in this paper so the question was when should an

entrepreneur decentralize using the blockchain the answer that I've given you this was the main result is decentralize if and only if the locked in effect of the users is sufficiently large and I had argued well then the commitment that the blockchain provides is valuable if there is this locked in effect that can be exploited and as kind of the very large of the very last bullet point that I want to give you is I want to name once again companies that have these decentralized counterparts and you can make your own opinion on whether or not the story makes sense and so I had AWS I had Google I had Facebook Twitter and Spotify so thanks for listening and that's the talk from my side and I would be super

happy about any comments and questions thank you [Applause] uh so maybe one questions this this is sort of something that Scott already brought up but I feel like there's some conceptual overlap between the rules of little U and capital V and you know right so Scott was basically saying like what if little you like capital V depended on the network size yes and so and and so on the flip side one thing that's a little bit unnatural about your you is this kind of Step function right like as soon as you join it goes from

something zero to something non-trivial whereas you certainly expect something more gradual so I'm just wondering if like I'm wondering if actually there's some version of this where the Primitive rather than having this sort of phase transition of you you want to characterize something which is basically like capital B minus U as a function yeah the network size and then somehow like now now it's not a one-dimensional thing anymore but you know what I mean kind of again that's sort of two sides any thoughts on that or yeah yeah so for the first part right you so I I kind of talked a bit about when Scott asked I can make you dependent on uh on the user size as well so if view is increasing and it I can

get a model that goes into very similar Direction and the reason is that you know at the very start when the network was small I didn't even have a commitment problem at all I was intrinsically motivated to grow right and if you was kind of small then as well and that doesn't change much and when the network kind of was growing here that's when I get the commitment problem and if now also the size of clocked in effect kind of gets very large there then that's just worse for me so that I can somewhat easily do and the second part about having this kind of Step function approach um like it's very nice for convenience to get the mathematical result right but of course you may think you know this builds up over time a little bit I think

the story will still go kind of through but the math would be a bit more complicated so I was thinking about you know maybe putting this into continuous time and having it build up over time and so on um yeah I haven't done it yet or I don't know if I will but I think it's certainly interesting um yeah but I think kind of the main story of the paper that you know I have this commitment problem over time I think no matter how you model it it's still going to shine through in the end so that's kind of why I went with a little bit of a simpler approach I had a question on the building side of um so I guess in this model entrepreneur

doesn't vote at all in the monetization rate but I guess like it's kind of natural to imagine in these Israelite systems of like the entrepreneur might like do some like simple thing or like somehow did it ended up also building um on the monetization ready like the procedures reporting too so you consider like if you add the entrepreneur to those votes um how the equilibrium changes yeah so for this model I can actually let the entrepreneur vote as well and why well the entrepreneur gives away half the kind of Revenue share of half the tokens and if I tie the voting to the tokens and I give away half and it's a majority vote and I'm actually not winning you know maybe I give away 50.1 percent right but in this equilibrium your users

they're all going to vote for the same thing the entrepreneur may vote for something different but the entrepreneur is actually not winning so I can do that pretty pretty easily in the model it gets more complicated of course when users are now heterogeneous as well and the entrepreneur has a very large voting blog and you know then maybe I need to give away a bit more or the users need to coordinate a little bit better then it gets more complicated but in this model I can allow the entrepreneur to vote as well it's not going to make a difference a quick follow-up on that maybe um suppose you did allow the entrepreneur to vote and they did have majority control or whatever presumably then they have exactly the same commitment problem they had before and so you get back to the essentially the centralized equilibrium is that the

right interest yeah exactly right there so interestingly here in this model the entrepreneur would endogenously choose to not have a majority voting stake exactly exactly I would give away more than 50 fascinating yep got it sorry you said that I just I hadn't parsed you properly no no I I should I should make that more specific uh probably also in the paper right because it comes out very naturally here but it's it's an interesting point so I I think it's good if I'm going to be a bit more specific there yeah do you have any ideas of how you would uh measure locked in effect in reality in these platforms um so personally I don't have much ideas

so I'm I'm kind of a serious so I haven't measured much but but there are papers and I I have them in in my paper kind of who actually measure logged in effects and I think some of those methods you could use also for these networks but for doing at the practice I would have to actually look into how how to do it but but the people have have tried and have thank you cool question if we have time um do the users all have the same discounted total utility regardless of when they joined um

yes right because in the growth phase at the very start I get zero utility right so kind of the only parts that matter is later on the um kind of compensation and the uh abuse after that or the exploitation but since you know after that nobody else joins as well everybody is actually going to have the same utility no matter at what point in time they join all right let's thank the speaker again thanks so much for having me it was a pleasure