DeFi Incentives - Running Your Protocol Like a Business - Gabriel Pohl-Zaretsky

huh yep hey guys super good to meet you all uh my name is gabe polzeretsky i work for gauntlet networks and today we'll be talking about d5 incentives and how to run your protocol like a business and not like a ponzi scheme so first of all my name is gabe i have a bunch of experience working high frequency trading for the past three years mostly at companies like jane street and other global trading systems both like sort of algorithmic trading firms i've been working at a company called gauntlet for the past uh year and a half basically using my expertise in traditional finance and try to apply those things that we've learned there to the more d5 market so most recently i've been working with sushi swap to build a statistical model to basically set their token incentives and that's what we're going to be talking about today um gauntlet as a platform is a financial modeling platform that uses sort of techniques that we take from the traditional finance sector and move it into informing on protocol on chain management to help inform d5 protocols about how to maximize both overall output as well as minimize risk so what are we talking about today so basically the biggest takeaway that i want you guys to take away from this is that daos are corporations like and we like to in crypto think that we're doing something completely separate and completely novel but in a lot of ways like what we're doing is been done before right so if you think about shares like you can call those tokens you think about vc raises ipos those are token emissions and icos shareholders token holders board of directors core team all of these things have corollaries in the real world equities market and we have to treat our businesses like businesses and if we don't we're not going to see the growth that we want to and know that we can't achieve so the question then is what do we actually have to think about what do we have to analyze to make sure that we're building long-term sustainable businesses that drive long-term revenue both to our shareholders as well as to our users of our protocols so the most important thing as a business is how are you spending money like you have a budget that you're currently allocating in our form it's token emissions so how are dows currently spending money so let's go through a couple examples so first of all we can look at curve finance curve finance has if you look at their overall token emissions schedule over the four years about 11.6 billion dollars in total token emissions that they're going to be spending over that time and where does that go so if you look at the pie chart you see about a little over half goes to liquidity mining incentives about 36 percent goes to shareholders and founders quite a little bit to community through like air drops and then you also have like an employee fund for new employees to be hired later on you see a fairly similar thing with balancer about 75 percent of their liquidity mining incentive go to liquidity mining incentives 25 to shareholders and again we're talking about a large amount of money about 1.5 billion dollars over the lifetime of their protocol same thing with dydx 50 towards liquidity mining 27 shareholders founders employees you see a very similar picture and again we're talking about a huge amount of money 7.8 billion dollars over the lifetime spending of the protocol lastly compound finance similar story 40 liquidity mining shareholders founders governance incentives employees almost half a million half a billion dollars over the lifetime of the protocol at current token prices so first of all we're talking about a huge amount of money and second of all you guys might have noticed the trends liquidity mining incentives are huge it's what most of these protocols spend most of the money on and so it's absolutely imperative that they get these mining incentives right and they're making the most use out of their incentives as possible so what are liquidity mine incentives to the uninitiated um this is something that d5 protocols use to attract capital towards things called liquidity pools so there's used in a variety of ways and a variety of different mechanisms so first of all you might have decentralized exchanges right units will have sushi swap curve balancer and they're using these incentives to drawn capital into automated market makers so that people can trade on these platforms you also have lending protocols maker compound ave liquidity doing a very similar thing except instead they're enabling people to borrow and lend crypto on these platforms you also have derivatives protocols dydx synthetics ribbon all of which are doing a very similar thing attracting capital attracting liquidity providers to make it so that their protocols can function and so this is incredibly important these protocols don't work without these incentives and they also don't work without this liquidity so that's why they put such an emphasis on these liquidity mining incentives in the first place so if you really kind of get down to it liquidity mining is an optimization problem so we can kind of make some assumptions we can assume protocols exist to maximize profit or some other kind of objective function protocols have a fixed budget to spend usually set by governance although of course this can be changed through later on votes and then protocols have a fixed number of pools in which they can actually allocate that budget so if you kind of formalize that right you can then just construct that as a mathematics problem of maximize expected profit as some sort of a function given an allocation of pool budgets subject to that the overall budget is constrained by the total budget that the protocol is allocated so this is the problem that protocols are trying to solve and currently they're really not doing a very good job it's a hard optimization problem so we can take a look at trader joe which is a pretty large dex currently on avalanche protocol and we can look at sort of what their liquidity mining incentives are currently doing so if we look at two competing pools you have usdc ustt on avalanche and then you also have usdc avax pool and if you look at these two pools you see that the statistics about these pools don't really seem to line up right so if you look at the stable coin pool you have 39 million dollars in liquidity 2.3 million dollars in daily volume which generates around six thousand dollars in fees at a daily reward of fourteen thousand dollars 14.

5 a thousand dollars so if you kind of crunch the numbers there that gives you a liquidity utilization of five percent and what that means is for each dollar in liquidity how many dollars in volume are you generating and then so the cost of that liquidity just by taking the total amount of liquidity and dividing it by the total amount of incentives that they're spending is 157 dollars of liquidity per one dollar in joe token so what this means is trader joe is currently spending one dollar in incentives to attract 157 dollars in liquidity and then if you look at from a revenue perspective in terms of the fees that they're generating they're spending one dollar in incentives to generate 43 cents in revenue now there might be good reasons for this maybe they're trying to drive growth maybe they value this pool for some particular reason but if you contrast that with a different pool if you look at the usdc avax pool on trader joe you see that this pool is by far more productive it has 163 million liquidity so it's quite a lot larger but for that liquidity they're getting 72 million dollars in volume generating 185 000 in fees at a cost of only 140 000 in rewards so the liquidity utilization here is 44 every dollar in liquidity drives 44 cents in volume so it's over 20 times that of the usdc pool and the cost of liquidity is 905 dollars liquidity per one dollar in joe so it's almost nine times bang for buck in terms of liquidity itself now if you think about the most important thing which is revenue joe is actually an avax pool earning a dollar and 30 cents in revenue in terms of fees for every dollar that they're spending incentives so just at a very quick pass it becomes very clear to anyone looking at this that joe would probably be doing pretty well to move some incentives from usd usdt into the usdc avax pool and so why aren't they doing this right so this sort of the issue of like how do you make these decisions how do you figure this all out so from talking to quite a lot of protocols about this um the way most of the protocols do this is essentially through gut feel so protocols have either sort of one quants with an excel spreadsheet or they're just kind of setting things based on like what they feel is right so the baroque approach gut feel you just kind of set what seems right the woke approach is you're kind of using more quantitative heuristics so you have some sort of an excel spreadsheet you're using some heuristics and while this works better than what we've seen for a lot of protocols it's still not quite exactly the most utility maximizing way to be setting these incentives so that's why we at gauntlet have been working on incentive optimization products and the goal there is basically to create statistical models to very quantitatively and very scientifically set these incentives across the various uh protocols that use them to maximize this impact for this huge amount of spend that protocols are currently doing so we can take a look at our model architecture basically what we're doing is we pull on a variety of on-chain data so we're pulling in data both from sort of like pool performance understanding what's going on in the market and then we also run a variety of sort of off-chain simulations and various analytic platforms we then download that into our internal data stores where we can build business intelligence dashboards that we can then show protocols to give them a deeper understanding as to what's going on in their internal pools and then using that data we can train statistical models both trying to predict the relationship between incentive and liquidity as well as the relationship between liquidity and volume so what that means is basically we're trying to predict what will happen if you increase incentives in a pool if we increase incentives in the usdc pool by 10 how much liquidity do we expect to inflow our outflow based on that change and that's a very difficult problem it requires us to understand lp behavior some lps are more elastic than others and more likely to respond to those changes and then after that we're also trying to predict relationship between liquidity and volume if you increase liquidity by 10 how much do you expect volume to increase by and this is always a moving target as market conditions change as the utilization of pools change some tokens will moon some tokens go down you always need to be readjusting these models based on real-time data to understand exactly what's going on that's exactly what we've built at gauntlet so once you have these two models it actually becomes a fairly simple optimization problem right so you plug in these two elasticity models into a customizable incentive spend optimizer where the protocol can come to us and say these are our goals with these pools in the most simple case maximizing overall volume in more complicated cases and might be making sure there's like a minimal amount of liquidity for partner tokens native tokens something along those lines we then plug these models into an optimizer and that enables us to do two things first of all it allows us to do what-if analysis where protocols can come to us and say if we were to decrease our overall incentives by 10 what's going to happen are we going to lose liquidity are we going to lose volume how much and it allows protocols to sort of answer questions about what's going to happen before they actually do it to enable them to make the best possible decisions that maximize the utility of the protocol the other thing and this is the more impactful thing we're able to do with this is actually generate incentive recommendations so this is what we've been doing for sushi swap for the past year where we've actually been setting the incentive recommendations and allocations across all of their pools and so based on that we can go on chain and actually execute those incentive recommendations set all the incentives for the pools and then after that we can then actually update our models based on what we view as feedback from that data so we may increase liquidity in a pool based on a belief that it's going to have this impact after making that change we will then ingest that data and understand how that change affects the pulled develop in real time the models will then ingest that data and update themselves uh accordingly and then the next set of recommendations that we're going to generate are going to be incorporating that data to make sure they're as accurate as possible so how well does this work um with sushi swap we've seen pretty strong results um based on our platform sushi was able to increase their efficiency by about five x uh what that means is basically for every dollar that sushi was spending in incentives we were able to increase the amount of volume that they were seeing for that by about five times and that leads to about 300 million dollars in daily volume based on the product that we're selling uh we've also been working with balance here on sort of fee optimization doing a somewhat similar thing trying to sort of tune their protocol to make things as efficiently as possible so the key takeaways here are first of all first of all that protocols are a business um they need to be worrying about revenue and spending you can't simply sort of throw money and do things totally uh ad hoc protocol spend a ton of money on liquidity mining incentives it's the most important things for protocols to be thinking about a systematic quantitative approach is absolutely necessary to be doing this you can't just rely on gut feel we're talking about too much money and too much spending to just leave it there and that's what we're doing at gauntlet so if you're a protocol that's interested in this product come find me um super hyped connect can also send us an email or follow me on twitter uh the other thing is we're hiring so if you're a data scientist or an engineer or a sales person who's interested in this type of product and would like to work with us come find me more than happy to chat as well but happy to answer any questions i don't know up here or later after the stage but yeah thank you guys so much for your time [Applause]