Edward Yu – Bringing OTC On-Chain and the VariationalOMNI Perp Dex (S7E18)

Hey everyone, Corey here. Thanks for tuning into another episode of Flirting with Models. If you're enjoying the show, I'd greatly appreciate it if you'd take a moment to rate, review, and most importantly, share with a friend. A word of mouth is how this podcast grows. And if you'd like to learn more about newfound's platform of Returnstacked Mutual Funds, ETFs and model portfolios, head over to Returnstacks.com. Now on with the show. All right Edward, are you ready? I am, yeah. All right. Three, two, one. Let's cham. Hello and welcome everyone. I'm Corey Hofstein and this is Flirting with Models, the podcast that pulls back the curtain to discover the human factor behind the quantitative strategy. Corey Hofstein is the co-founder and chief investment officer of newfound research. Due to industry regulations, you will not discuss any of newfound research's funds on this podcast. All opinions expressed by podcast participants are solely their own opinion and do not reflect the opinion of newfound research. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of newfound research may maintain positions and securities discussed in this podcast. For more information, visit thinknewfound.com. In this episode, I speak with Edward Yu, co-founder of Variational. We begin the conversation with Edward's background in cryptoOTC markets. He explains how the space evolved away from telegram chats, the complexities of pricing derivative structures on the long tail of alternative cryptocurrencies, and the sources of natural flow in the space. This experience led Edward to co-found Variational, which seeks to bring the trillion-dollarOTC derivatives market on-chain by disaggregating settlement, margining, and derivative payoff logic into programmable primitives. Built on top of Variational is the Omni-Purpedex, or decentralized perpetual futures exchange for the non-crypto-speaking listeners. Unlike other Purpedex's that are built around a centralized order book, Omni effectively acts as a user interface to an OTC RFQ system. On the other side is OLP, the Omni-Liquidity provider. This structure allows Omni to provide significant depth of liquidity on a huge breadth of investable assets despite the platform being in close beta at the time of recording. Given its unique design, we spend a significant amount of time discussing the pros, cons, and risks of this structure. This conversation is obviously outside of my usual realm, but for those listeners interested in market structure and where the world of finance may be headed, this is one not to miss. Please enjoy my conversation with Edward you. Edward, thank you for joining me today. Excited to talk about something very, very different than my usual fare for this podcast, but I think the listeners are really going to enjoy this fun, so thank you for taking the time and joining me. Thank you for having me, I'm quite excited to be here. I want to start with your personal path into the world of crypto. You originally studied applied maths at Columbia, you started with trading FX, jumped into stat-arb in crypto before Purps were even really a thing. Talk to me about what was the early edge that you saw in the crypto space? What made you move into crypto so early? What gaps were you seeing that maybe others weren't? This was basically 2017 and earlier, before 2017 I was just an undergrad at Columbia, and my personal passion back then and still to this day is basically in Bayesian statistics. We were trying to find a way to make that useful or applicable, and so naturally we wanted to try our shots at quantitative trading. The first thing we actually started out with was FX trading, but we didn't really get very far there, and at the time crypto market was developing, and immediately we saw, okay, there's 100 times more opportunity here. It's growing so fast, it's very exciting. So right after I graduated, we started a small trading firm called Q Capital, and we raised a little bit of institutional capital for it. One of our backers was Digital Currency Group DCG. To go back to your question about what was the early edge, I just had to give some context around what the market was like back then because it was truly just really crazy, really wild west. First of all, if we look at a factor decomposition of returns back then, right now market data is about 50% of variance or so back then it was 80% of variance. Naturally, a lot of the funds out there, they were just long only, and that was very smart, and those are the ones that really made the most money in dollar terms. We were doing something different, we were doing one of the first either long short or you can call it basic stat art stuff. Other than the 80% returns are explained by just market data, what is left in the 20% that is interesting? And at that time, it wasn't even that easy to short all these arc coins. So for example, we were onboarded with this essentially CFD or like OTC desk called B2C2. I think they're still around, but back then when you were trading CFDs, you wouldn't know what the funding rate was until after the fact. So they would email you every day at like 4 p.m. and you would get an email and say, okay, these are your positions and this was our back adjusted funding rate and you paid this amount. A lot of the game was not only trying to predict the market movement, but trying to predict how crowded the trade was from other market participants because sometimes the trade itself would be fairly obvious, but you would end up losing money because it was too crowded and the funding rate didn't go your way. So that was one of the things. Another thing was these perpetual exchanges were just starting to come online. The market structure back then was just truly insane. You could sign up for centralized exchanges and they would give you some 50x, 100x leverage. No KYC. The equivalent volatility was averaging triple digits, something crazy as well. So some of these exchanges were not good. I remember on BitMex at first, their liquidation engine was not what it is today. So some people were like, okay, if we just go insane leverage and just get liquidated on purpose, they couldn't really collect on the negative balance. That was some one of the asymmetric bets that people were doing back then. Just various stupid stuff like that. It died pretty fast, but it was generally a fun time and a couple years later the market fell out and essentially recovered from there. Your experience in those early days of connecting with OTC eventually saw you get brought into Genesis to "electronify" their OTC flow. So I mean, you could spend a little time talking about what the OTC landscape was like at the time and what your team did to change it. So Genesis has been around since forever, even way before I joined. I believe they started out in OTC for Bitcoin in 2013. That was completely chat, voice-based. You would hit them up on Telegram and say, hey, can you give me a quote for X amount of Bitcoin? And they would just charge you like a multi-hundred basis point spread, so a multi-percent spread. That's the core original business and they sort of expanded from there. By the time I joined, it was already much more sophisticated. So in the 2017 era, I believe Cumberland had already existed in spun out of DRW and then we were connecting directly direct private lines to all the market makers, Jane Street, you name it, the big names you hear today, they're all getting into crypto. So by then it was essentially already pretty electronefied. Our job was to compete with these desks for client flow but also sometimes work together with them if we need to offload risk. And this was basically all done through streaming RFQ system. We had a couple of DMM deals with exchanges as well, so we were active on the centralized exchanges at the time. But you could think of the majority of the flow in the business as being, there's various people that just want to trade systematically with you so these could be anyone from ATM providers, those Bitcoin ETFs, retail brokerages or ETFs, etc. Or I know with individuals, just all of that flow coming in and our job was to figure out what to do with all of that and so forth. In our pre-call, you described your edges being ultimately in monetizing client relationships in the OTC world. Can you unpack what that means and maybe talk about the sort of client flow that you were seeing? In theory, it's not hard because you have these sort of clients. Here's what I call natural flow. They're not making their money on speculation. They need to do things with crypto. So for example, this was before US ETFs at least, but think of that sort of thing. They have systematic buying and selling. So once you get that, it's not so difficult to figure out what to do with it, but the business is very competitive. If you give them a bad spread or bad execution, then they're going to like three other deaths and getting similar quotes. Your job is to say, okay, how competitive can we go against jump our Jane Strait or whoever they're also competing with and still make profit on that spread? In your experience and time, you've seen within crypto OTC evolve from telegram chats to now and we'll get into this with variational full stack RFQ protocols. What do you think is the most misunderstood thing or the big thing? because misperception people have about cryptoOTC. - One thing is there is a particular R&D FI where all these quants are dreaming up all sorts of weird products. We had like ETH squared, which would give you like returns of ETH but squared. And then people were pitching all sort of like exotic options. Oh, do you wanna buy something where it's the best of Salana Ethereum and Bitcoin performance over like a certain time period? People were coming up with all sorts of crazy things but I think at the end of the day, the markets need to be driven by natural flow. One analogy I have is agricultural options or futures. They have a very natural base of farmers and related parties who need to do certain financial transactions and their primary job is not speculation. And that sort of creates all these different dynamics around the market. For example, they have seasonality or volatility risk premium, whatever. That natural flow is really the lifeblood of all these markets. If it's just a bunch of smart trading firms, PVPing, each other, that's very tough game. And I think the volumes usually sort of dry out. So when we think back to crypto about who the sort of natural flow is, I think it's a few different areas. One, it's people who actually create these tokens. So projects, founders, foundations, investors who deal with that sort of thing, VCs and so on. And then you might see some investor demand from retail as well. At the end of the day, all OTC products, they exist to serve that market rather than weird PVP things. - I want to talk about that natural flow a bit because I can imagine a genesis, a lot of the OTC flow would be pricing options and probably founders and treasury selling call options on obscure altcoins, some sort of structured product on some obscure altcoin. Can you talk a little bit about what the hardest challenges are in bringing these structured products and derivatives infrastructure to this massive long tail of crypto assets? - I think that was something really unique about genesis. And to my knowledge, that particular market has never recovered. I think most volumes these days for crypto options are concentrated in the top few majors. I know there are OTC tests serving the rest, but to my knowledge, it wasn't as active as that crazy bull market, that 2020-2021 era. It's really interesting and a really tough problem to make markets around these things, which is not your usual step. Because usually at least you have a very easy way to get a baseline model for what we're doing today. I know what the fair value Bitcoin roughly should be. I can look on 100 different exchanges and average them. But for options on altcoins, it's really not like that. At best, you have the BTC and ETHVOL surface. And then outside of that, you just have nothing, no visibility whatsoever. So everything is all in-house and all proprietary. Luckily, it turns out if you have good flow, actually vanilla black shells works pretty well for this. Even if you have a flat-vol surface, but you have pretty widespread to start, it can be pretty profitable already. And then you're watching the flow come in and you have a bunch of smart traders that are adjusting things as people head on various areas of a VOL surface. And so that's really interesting dynamic to essentially start from zero and then evolve it in-house to what an official market should be. On the quants, IB basically just provided baseline models. So we were saying, OK, given where the BTC VOL surface is trading, this skew is this much. We think a good baseline model could be this for this altcoin. But there's certain dynamics that are very specific to that altcoin. And then it's really the trader's job to have a bunch of domain knowledge and to manually adjust these things for their clients for better pricing. So we just did the baseline work. But a lot of the credit actually goes to the traders themselves. And they have their own techniques for doing that. You look back at your time at Genesis. What do you think were some of the most impactful projects your team undertook to help modernize and electronefy this OTC flow? Genesis had an interesting path starting all the way in 2013. As it evolved, they were sort of these disparate flows or disparate desks across the world. So we had a desk in Singapore and London and other cities, et cetera. One interesting thing we actually did was we built an internal-- you could call it a dark pool or an exchange. Functionally, you could think of it as a coin base. It was an entire full-fledged exchange. But it would only be available to internal Genesis traders. If a trader saw something coming in one way, he could basically submit a limit order or submit a T-Wop or submit whatever algorithm he had to this internal exchange. And our market-making systems would pick that up and provide liquidity against it. And this would also inform whatever quotes we were doing on the LITS or centralized venues. This is our way to basically connect the global company into one cohesive system. And it would give you really nice, just reports and accounting average fills on the internal exchange versus external and your markouts and whatever. It was fully functional in that sense. We're going to spend the vast majority of the rest of the conversation talking about the protocol you're working on variational and the purple decks that currently lives above it. Can you connect the dots for me on how you got there? You're working at Genesis. You're working on the OTC desk. What was the inspiration to leave and start variational? And can you provide sort of a 30,000 foot of view of what problem you're trying to solve? Timeline Wise, we left Genesis in 2021 and found a variational. The very first iteration of variational. It actually started out as sort of a trading firm as well. At first, we were market makers in very particular sets, which was our niche, which was generally providing the liquidity to defy protocols. And also, we did some ongoing option stuff. They had brokers that would aggregate flow in seven to us and we would be at the other end of that quoting them. As we were building our business, we actually stumbled upon a much bigger opportunity, which is to sort of build that tech side of the OTC desk. At its core, what variational is, is it's a peer to peer way to bring any derivative on chain. And it relies on as few externalities as possible and as as customized as possible. So that's what the protocol itself does. And we really started it to solve a lot of the problems that we were seeing in the OTC space ourselves. I don't know if most people have gone through this experience, but it's kind of a pain to get started with the OTC desk, especially in derivatives, because what happens is, first, you need to get a bunch of lawyers to negotiate the ISDA agreement, which is a few hundred pages describing exactly what happens to collateral and what happens in the event of a default and how they're marking everything to market all these nuances around how these OTC derivatives will trade. That takes a lot of work just to get onboarded. And then it's pretty manual after that too. Usually you hit up OTC desk via telegram chats, you're like, hey, can I get a quote for XYZ structure? They reply back and then you say done. And then settlement and all this margining usually is done manually as well. So you're sending funds on chain back and forth to each other to settle up margin and so on. We thought all of that was just very inefficient, very messy. So in particular, we wanted to build tech for OTC desk that could automate that entire pipeline. Our vision was like, hey, what if we just gave you legal blocks for derivatives? So you could define a payoff function for a derivative. You could define what Oracle or Time series you want to use as the underlying. You could define different margin, liquidation rules, and how settlement would work and so on. And then both of you would connect your wallet, deposit a collateral, and you'd have a peer-to-peer traded derivative just end to end. So that was sort of our original vision for things at the protocol level. It's very generic. So we're sort of building consumer-facing apps on top of that, one of which is called Omni, which is for specifically perps trading. You can think of it. The experience very much like it perps decks. And we're building another one in the future called Pro, which is for OTC desk. If someone's like spin up their own OTC desk, they could do it with variational Pro. And our pre-call you described variational, as, quote, Robinhood plus Citadel of crypto. Can you walk us through what that means from a user and platform perspective? I mentioned we have this app, which is built on top of the protocol called Omni. And the user experience is very much similar to a Robinhood. And what I mean by that, it's basically an easy way. A few clicks, you're signed up and you're trading long shorts, 500 different perps. Anything you can think of is covered. So that's the Robinhood side of things is the UI. What Robinhood does is they basically route your flow to market makers directly, skipping some of the, I guess, in crypto, defaults, limit order books that people generally trade on. So the parts, which I mean about how we're building in in House Citadel, is we've got a liquidity provider called Omni liquidity provider, OLP, which is providing the liquidity to every single person. single trade on the Omni platform. So you're using this app, you're trading, and then all of that is getting routed basically to one liquidity provider. The upshot of that is basically that our revenue per dollar traded is 2x or competitors, because you can think of it as we're vertically integrating. So we're capturing the market maker side and we're capturing the exchange side. So where we see value is we want to find a way to revenue share that back to the user. So return that somehow in the form of various discounts or rewards or zero fees, etc. I want to stick with variational firm and before we talk about Omni and how they're integrated, but the core variational protocol design centers on this bilateral OTC style settlement pool. What makes this architecture uniquely suited to deep liquidity or fee elimination and ultimately scalability of product solutions, especially when you compare to the solutions that are out there in WebThru right now, AMMs and traditional order books. So the key thing to understand, as you pointed out, when you turn on variational, it's peer to peer. You're directly settling a trade with your counterparty and both of you are posting margin and that gets locked into an on-chain escrow contract. So the risk model is basically the opposite of a centralized pool that you can think of like an exchange. And an exchange model, you don't care too much if your counterparty goes in solvents. As long as the exchange itself is solvents, they will cover you and they have insurance funds and all sorts of different mechanisms. So you don't have to care about that sort of thing. However, on an exchange, you do have to worry about the solvancy of the entire exchange. It could be that you do something that's not your fault at all and the exchange blows up and everyone on there loses all their funds. You can think of the peer-to-peer model as the exact opposite of that. You never have to worry about the protocol itself going in solvents. It's just code. The protocol itself is not providing capital, etc. But essentially what you are worrying about is your counterparty going in solvents. So it's a very different sort of risk model with its pros and cons. But one of the huge benefits of this is that there's no financial contagion. If there's some sort of error or something wacky happens and you manage to create like a trillion dollars worth of bad debts on an exchange that sort of socializes across everyone. But in a peer-to-peer model, that's stuck essentially within that isolated pool. So it doesn't spread to other pools which are isolated from that. One of the benefits of doing that is it allows for more long-tail trading. Stuff that could be a little bit risky for a margin engine to be listed on an exchange because it potentially affect everyone. But this gives more flexibility on what market makers are willing to accept because again, it's super isolated and you could have different market makers specializing in different sort of products for that as well. Back at Genesis as we talked about before you were doing some pretty complex OTC structuring your pricing options on obscure altcoins. I think a big part of that was locked tokens being delivered as collateral, which creates a whole time-value element of what you have in your collateral. How did your experience there inform the underlying design of variational and how does the protocol make those types of bespoke messy deals tractable for OTC desks today? One thing that Genesis was very good at and I core part of their edge is essentially understanding the business need first. All the quantum modeling and stuff is layered on top and goes after. But at the end of the day it was the client who said, "Hey, we have a bunch of locked tokens or we haven't agreement to buy future tokens etc. We need to do something with this. We want to generate yield on this, whatever. What products do you have that can help us?" And that's how that market segment came about. From that, I think one lesson that we can learn is we just want to give legal blocks. We don't want to be particularly set in stone like, "Okay, you have to do only this type of trade or we put a bunch of restrictions around it. So just make it as flexible as possible. You've got this settlement pool. Do you want to bring your own collateral? We want to accept x-type of tokenized agreement or we want to accept y-type of asset as collateral?" "Okay, that's up to you. We want to import our own market prices. We've got this stream going. Okay, as long as you guys trust each other, go out it, have your own Oracle and we'll just provide the infrastructure to automate, margaining, liquidations, and so on. But everything in the protocol itself is customizable." You've effectively disaggregated settlement, margaining, and pay off logic into these programmable primitives, these legal blocks as you're calling them. As you envision it, what kind of flexibility does this create for institutional OOTC desks or what sort of new use cases can you envision for this sort of structure? "Our dream, a really long-term vision, would be a bunch of segregated financial networks. One analogy you could think of as Shopify for derivatives, we wouldn't even know the extent of what people could do with it. It's truly up to people's creativity and imagination what they can do with it. But I'll give some examples. Each segregated network could have its own market that it specializes in. For example, we built Omni, which specializes in crypto-perps. So that's one type of market. Someone else could spin up one where it could just be perps on equities or perps on prediction markets. It could be local. Someone could focus on Hong Kong stocks or South African stocks, etc. The point being that everyone can create their own financial network here. It would be efficient and people could come in and compete with each other. Really, we wouldn't necessarily put so many restrictions on what people are allowed to do. It would just be completely open legal building blocks for that." "I want to switch to talking a little bit about Omni because I think it's a really interesting use case of the variational primitive layer." Omni, as you mentioned, is the currently retail-facing perps platform, though I guess any institutional that wants to trade perps could tap in there as well, while variational is the underlying protocol that powers it. Can you walk us through a little bit about how Omni is ultimately built on top of variational and how that separation between protocol and product shapes what is possible on Omni and maybe what makes it unique from other perps dexas? "When we were building variational and a really long-term vision of Shopify for derivatives, we realized that wasn't super realistic in the near-term because it was just giving users too much choice and too much legwork. How are users supposed to go source their own liquidity and their own financial networks and spin up everything themselves? I think most people don't want to do that. Even though this is still our long-term dream, we realized we had to do 99% of the legwork ourselves and make it just really dead simple for users to use." Omni is a first-class app on top of the variational protocol, which means that theory anyone can spin up their own Omni clone using the same protocol. We use the exact same primitives as anyone else when using this protocol. When users sign up for the platform, this actually spends up a peer-to-peer settlement pool between the user and the Omni liquidity provider OOP, the same margin liquidation and gen. We have predefined rules around that that are set very specific and we set the derivative payoff type to be perpetual futures, etc., etc. But theoretically, someone else could come in and say, "Hey, we're a developer team. We want to build something similar. Can you help us?" And that would also be possible. As far as I understand it, as I believe this is what you just said, every trade on Omni is effectively an RFQ to the backend OTC protocol that's getting filled by the OLP. This isn't your traditional order book market maker structure. This is actually a Perp Dex where you have one major counterparty on the other side that is fulfilling every trade as an RFQ. That is a massive difference between Omni and every other Perp Dex I've looked at. Can you talk about a little bit why this structure matters and what it implies for whether it's retail or institutional participants on the platform? I think the RFQ is basically a natural extension of the peer-to-peer aspect. When you're a user, you actually have your own settlement pool versus OOP and another user would have their own isolated pool as well. At the end of the day, OOP is managing thousands and thousands of these settlement pools, the treasury management system that's moving funds in between all of them as well. Naturally, if it's just a two-person market basically, I think that's doing a full fledged HFT style limit order book is a bit heavy. The lightest touchway that we could do this, which gives you good liquidity, is the RFQ model. As long as the user tells us what quantity of something they want to buy, OOP can give them a pretty tight quote on anything and then there essentially one clicking their way through. It'd be pretty seamless for them without us having to manage thousands and thousands of order books on the back end. Most Perp exchanges monetize via spreads and fees. I'm not as if I'm misquote you here, but I believe you said the OOP structure gives you basically seven dips for every dollar traded, which is if I'm getting my numbers right, probably more than double what you see with most Perp Dexes and you're not explicitly charging users. Where's all that extra? for an inch coming from. Users are also sometimes a little bit surprised because I say, according to our market research, we're making 2X, the amount of revenue per dollar traded. And for the user, the transaction costs is actually lower compared to like, buy and answer hyper-liquid, etc. I think for exchanges, they actually do not capture the spread. They only capture on fees. And it's the market maker who earns the spread, which can be a little bit of a shame in some assets because it's not always an even split. If you got some super far out there, all coin, and just 100 basis point spreads, the exchange is still only getting their 3 basis point fees or whatever. And 97 basis points is going to the market maker. It's not exactly free for the market maker because, ostensibly, they're hedging costs and pricing costs are larger for those all coins as well. But there can be some markets where that, it's dramatically shifted in favor of the market maker for some of these. And for some markets, it's more in favor of the exchange in terms of profit margin. The key thing is, by vertically integrating, we're capturing both sides of that. So it doesn't matter if the split exactly between spread or fee, it's just mostly about your top line transaction costs for the user. And then for us, it's about whatever top line revenue that we're getting. How is the OLP able to offer such deep liquidity and such tight spreads? And I'll just use an example from my own experience, which is that I have quoted millions to tens of millions of dollar trades on hyperliquid and Omni. Omni is still in private beta with very few users, very little open interest. And in many cases, Omni was quoting a tighter spread for me than I was getting a hyperliquid. How is that feasible? What's actually happening in the background with Omni to make that possible? The core thing is basically solving adverse selection problem. I think that the core function of a market maker should be to provide liquidity for natural flow, for assets. But in the real world, what usually happens on exchanges is that's only a small piece of the puzzle. And most of it is actually just about PVPing other HFT firms. Sometimes people are complaining because, okay, spreads look tight and it's a bunch of HFT firms on there. But they actually have no obligation to provide real liquidity. So when things go a little bit daywire, they can simply just pull other quotes and the market is not actually liquid. So I feel like that market structure is getting away from what a market maker is supposed to do and delving into all this HFT arms race stuff, which isn't necessarily mostly about providing liquidity actually. I guess one good thing about OOP being the only liquidity provider on platform is that we could just get back to the basics of that essentially. There's just user versus OOP providing liquidity. There's no weird HFT games like hiring FPGA engineers. None of that. Getting away from all that is where the cost savings are coming from and therefore providing benefits for users in terms of revenue sharing for them. Typically you might expect spreads to be tighter where there is a lot of offsetting open interest with Omni. You don't necessarily see that. How is the OLP operating in the background to offer such tight spreads? And the background you can think of OOP as basically a standard market maker actually think of like a winter meter or something. It will quote you on trades and then when you settle the trade, it will be booked into the pool. And then on the back end, OOP has its own funds and important distinction. So it has its own funds sometimes on centralized exchanges for hedging. It doesn't place the user funds. So OOP is essentially posting collateral on the Omni pool and on the centralized exchange. Sometimes it will hedge the risk for user trades. Another key piece of the puzzle is that it doesn't just go and hedge user trades one to one on the centralized exchange, which would basically make it some sort of order router. I know that's a business model that some have built in crypto as well. But that's not exactly what it is. It actually holds inventory risk. It's our algorithms that determine, okay, is there too much tail risk here? Should we offset it? Or do we hold this inventory for a while and let natural just offsetting flow come through? If it's good to a very natural flow, this is basically offsetting a lot of the hedging costs because we're not hedging one to one. We're only hedging a portion of the flow that comes in. And again, all of that is what makes the spreads very tight. I guess one of the potential risks there is that the OLP stands behind all of Omni's liquidity. How do you think about that risk and maybe the ultimate capacity of the OLP, making sure that you have enough collateral posted to the user-selement pool, but also the collateral you need on the centralized exchanges? And ultimately, if you're removing the ability for other market makers to come in and act this counter-party, how do you think about the incentive? Alignment of that structure? I do want to clarify one thing, which is not that OLP is backing every trade-on platform. So one scenario, which is that if the OLP goes to zero, there's a hack or something or the exchange loses OLP funds, the users are still solvent because they've got their own money locked into the Omni pool, which is not on the exchange. So they could withdraw their funds. OLP would not be able to process future, essentially, unrealize P&L. They would not be able to credit that to the user because OLP has zero money. But it isn't true that users are taking, for example, risk on centralized exchanges. It's basically like trading OTC with the winter meter someone. It's a very similar model to that. Now I believe the second part of your question is about incentive alignment. I do believe it's very important that the vault itself, the LP profits are essentially public and also owned by the community. I think that if it were the case that the team was owning the LP profits exclusively, there would be some sort of perverse incentive, maybe, to trade. This would really be one-to-one PVP user versus OLP, and that the more money the user loses the more money OLP would make. So we want to get away from that model and have the team and the protocol make money in a more neutral way. What happens is that the team, of course, has expenses and it is a for-profit venture. The protocol, it doesn't right now, but it intends to take a fee based on the volume and spread. It's essentially like a payment for order flow model. Similar to a Robinhood. This is actually their exact model. We imagine that will be roughly around the same that Robinhood has. So about 30% of that will go to the protocol. Then the rest will either go to OLP, we will make the vault public so anyone can just deposit into that. Or it will go back to the user in form of loss refund or referrals or whatever incentives we have there. What about capacity? Let's say you take OLP, you turn it into a public vault. There's X $100 million there. You take on the out of the private beta, you make it public and suddenly there's this massive demand for liquidity that OLP can't meet. What are the offsetting forces? What does it imply for funding costs? How do you make sure that you have the right market forces designed in place that liquidity would come into the OLP to support the growing needs of the platform? I think that is actually one of the downsides of this design is that it doesn't support infinite capacity. It's basically limited by the capacity or the funds that people deposit into OLP. So sometimes there are cases where there's an exchange and someone just puts like a singular multi-billion dollar order and it's no problem because as long as there's a query on the order books, other people are filling that and that's all fine. And that wouldn't be the case with OLP. We do have limits. So if there's not enough essentially funds in OLP to cover that, you would essentially just be blocked from placing opening a large enough trade against them. This is again similar to trading OTC desks. If you're like, "Okay, I want to put a massive derivative trade against JP Morgan or something to like, no, we just hit our risk limits on this particular thing. We can't hedge anymore risk." It's similar to that. So it truly is whatever the community decides to provide is how much capacity everyone gets. That being said, I think generally it's a pretty profitable business model to be OLP. So I don't expect there to be a shortage of demand for that sort of thing. It should be proportional in terms of the volume that it's getting. So I don't think we'll run into limits for most users, but that is something it could happen and I freely admit that's one potential downside of this sort of model. How do you think about designing those risk limits to that risk budget, right? I think of the reason case with hyperliquid in James, when just comes in and opens up a massive Bitcoin long or something like that. How do you prevent someone from just eating all the capacity that the OLP has to offer as a single user? One thing that's always puzzled me about these D5 faults is why they don't hedge essentially. They're just one to one against the user's position. So people were praying for this guy to lose a bunch of money because I would go want to one into HLP and some other D5 Volta like this as well. I don't believe that design is actually super sustainable because one string of lucky events could essentially cause the Volta's take pretty substantial losses. With our design, we do have external hedging. So someone comes in with a $1 billion Bitcoin long or hedging it on other exchanges. We don't particularly care if a user makes our loses money in medium frequency sense. So in that way, I think that's also less adverse for the user. But yeah, if there's not enough collateral in OOP, when they try sizing in, they're going to hit some sort of like, hey, you hit the OOP max limit. You're going to be blocked. What happens in the case where the OLP strategy itself underperforms? There's a certain expectation here that your team is sophisticated and designed a good strategy, but playing maybe devil's advocate. Let's assume for whatever reason you guys don't do a good job. How would that ultimately impact the user experience for users of Omni and ultimately the protocol integrity? First of all, I do expect the strategy to do pretty well. Just because we've got pretty experienced guys on the team from market making firms both in TriedFi and in crypto, the sort of thing is well understood. I don't believe this will be a likely case. But let's say it is losing money consistently. I think what would be natural is people start withdrawing their capital. And then we sort of rage a market equilibrium where the capital in OOP is proportional to what people are willing to provide for takers. If that starts declining, then people on the takers side will also start hitting their risk limits sooner. The whole thing will just naturally scale down from there or up if it is doing well. So there's a sort of natural buoyancy or decreases if there's not enough demand for that liquidity. Given that the OLP and Omni is effectively built on variational Omni sits on top of variational, there's an RFQ, the OLP fills the RFQ. Is it theoretically possible that another high frequency OTC desk could step in, not a traditional market maker perhaps, but someone could actually come in parallel and compete with OLP for that flow as an OTC desk? Absolutely. We want to create a sort of open market. If someone else wants to make their own essentially OTC desk system and they have a website like OTC.marketmaker.com and you just connect your wallets and you can get quotes from them instead of OLP. And if they provide better, you could just trade with them. And that's all cleared on variational. I think that is a plausible case. And I think we would welcome that actually. I want to take a big step back. We've been digging into the details of variational and Omni, but I want to talk about broader design here. And your views on, do you think crypto is ultimately converging towards traditional market microstructure, which I think we're seeing in many ways with what are being called this summer, the CLAB-ORS, the centralized limit order book, or diverging into new paradigms. And obviously, would love your view on how variational fits into that. Nothing against CLABS. We do love them. I think they're a very natural price discovery mechanism. And I think there's always going to be a demand for at least to one venue where it's just super high frequency and super focused on the fastest news being priced into that venue. And that's probably going to be a CLAB. We ourselves couldn't exist without them, because I didn't know day, our fair values are also coming from these different exchanges as well. And we're providing a sort of different structure, but it's complementary rather than replacing. That being said, I think that a lot of the traditional market microstructure, for example, in equities, evolved by accident just over the years and is not necessarily the best user experience, or even necessarily that good for society. Of course, we do need some way to price and information fast. But what is the limit of that? And at a certain point, does anyone actually care? Billions and billions of dollars going towards FPGA's and microwave towers and whatever, just to get updates, one nanosecond faster? Is that actually improving markets in any measurable way? My personal opinion is no. And I believe a lot of people have also spoken out about this. Our focus is how can we bring new market structures that are more focused on the user experience aspect of it, rather than just getting into the old school arms race and all that HFT stuff? You mentioned that Omni and OLP itself relies on centralized and decentralized exchanges for it to be able to offer the user experience. It does-- is that implied that there's an inherent limit to the size that Omni can be of total per market share before it becomes the snake eating its own tail? You need these other markets to be sufficiently large and competitive for you to continue to create a better experience. But by creating a better experience, you cite in liquidity away from those places, which inherently creates a worse experience, because you can no longer hedge. I don't think there's a limit, actually, because maybe one misconception users have is if I just do deals OTC, there is no market impact. Of course, that's not true. Even if you do a large amount of all coin options OTC, we've got our Delta hedging program running. That's connected to centralized exchanges, and that's doing its thing, which, of course, will go back around and create price discovery. I think that's a similar mechanic for Omni, which is that if everyone is trading using this RFQ system, but on the back end, we're still aggregating, and we're still hedging, and that's still creating flows in centralized exchanges. So we're still participating in price discovery in a roundabout way. So it all is circular, and markets are efficient, and converge to that. One of the big debates that I see online is around the core tenets of decentralization. There are people who believe that if you're going to be a decentralized exchange, everything should be coherent in the decentralized philosophy. And yet, what you see with some of the very successful perp dexas is that they're a hybrid between decentralized and centralized on chain versus off chain matching. And our pre-call, I think you said, at the end of the day, users don't actually care. How do you think about balancing pragmatism with frankly idealism in protocol design in this sort of decentralized finance world? I think as it currently stands, very few protocols are 100% decentralized. But I do think that is a good end state for things to end up. So I think it's about taking at one step at a time and focusing on the meat of the matter, what gives you the most bang for your buck? So I will say, for example, your funds being self-custodied and locked into escrow contracts on chain is way, way, way better than sending your funds to essentially someone else's private wallet, which is a centralized exchange. And they could do whatever they want with it. So I think that's a big step forward. Another thing is should we have proof of reserves on chain that everyone can see? I think that's also a big step forward that maybe we'll publish at a stations once in a while from some old school auditing firm. That's a bit hard to track. These things are sort of big wins in the decentralization. Some things I believe are not so clear if they're wins for users and maybe people on the fringes care about them. But it's way too complex for anyone to meaningful be parsed. I see online a lot of people at the bait. So Salon is not as decentralized as Ethereum because a lot of their validators are in one location or some of them belong to one entity or whatever. And I think that's too far downstream from what users can parse for them to have a meaningful opinion about that. Even as a crypto founder myself, I don't necessarily think I would know enough to apply it on which L1 is truly more decentralized and what tail case scenarios can happen on L2s, et cetera. But focusing on what we can and getting core wins about decentralization is I think a good step forward. - We have come to the end of the episode. And I want to ask you the same question I asked, I guess at the end of every episode, which is outside of work, what are you currently obsessed with today? What hobby, idea, music, show, book? What currently has you captivated and captured? - I think I mentioned at the beginning and undergrad, I was very evangelical about Bayesian statistics and I still think to this day, especially in the AI era. One experience I had was using these different AI tools and they told me a fact and it seemed very plausible. So I didn't go double check it, but it turns out the AI completely made that fact up. I guess I'm very passionate about all models having confidence intervals. When the AI tells you something, it should also tell you what percent probability it's sure about something. And that probability should be calibrated and not also another made up number. Unfortunately, the field of Bayesian statistics has fallen out of favor, at least in the ML or AI community, just because it's not so practically scalable. But I would love for way for people to find out how to bring that sort of things back. So we can have uncertainty quantification and knowing what you don't know and knowing what you know. As someone who has been lied to on multiple occasions by chat GPT, I absolutely love that answer. Confidence interval would be a wonderful thing. wonderful thing, though, given how confidently chat GPT lies to me, I'm sure it would lie about its confidence interval, it will. So Edward, look, I love the conversation. Thank you so much for joining me. Congrats on all the success so far, and I really look forward to watching how variational and omnie evolve in the future. - Thank you for that conversation. It was a good time. (upbeat music)