Transcript
Speaker: Fibonacci pattern or something, which to be clear is a scam. I think at this time, Jump was still using crypto for like their interns. They were like, you could go trade here for a little bit and then maybe figured out that like we'll let you trade something real. Natural suspicion had gone out the window for me and we had like a third of our book parked on FTX. We had a balance sheet with hundreds of millions of dollars and Hyperlux was just kicking our ass.
Speaker: Welcome to a new episode of the InSilico Terminal podcast. And I once again got jump scared a little bit like in the last episode because I'm not used to when i when I have a guest that has a camera on, it like puts their their face on full screen now in in the software for me, which didn't happen before. So it was a bit unusual, but... I guess I'm slightly getting used to it every so often again. I'm appreciative of another guest that is showing their their face to me. and not just I'm not just talking to a black screen. My guest today Jan Walth.
Speaker: Did I pronounce it That's Thank you for having me here. You pronounced it perfectly. Perfect. You can call me Matt. I see. you Okay, awesome. um You've been in crypto for quite a while, right? Since like 2015, from what I've read?
Speaker: That's right. Longer? Man, it feels like a long time too. I'll say that. How about yourself? When when did you get into crypto? I got into crypto in 2017. And then um I started of like buying the first coins and whatever. I i did i didn't really ah go into it that much from a trading perspective. I was very focused on the tech. I really thought, like hey, this is going replace ah the monetary system or whatever. And then after the 2018 crash, I was like, yo, this is actually all a lot more useless than I thought. Because Bitcoin was expensive, Ethereum didn't have anything on the chain yet. And then I kind of left and I came back in like 2021. And since then I've been very engaged and then eventually it's now my my full time job and my life. But how but how about you? How is your crypto journey going? How did come about?
Speaker: I think we had like almost opposite entries because for me, 2015 through like late 2017, I was very focused on learning how to build a high frequency trading system. Like i was a software developer by background. i got roped into this kind of like GarageBand HFT project.
Speaker: um And so I was, you know, trying a million different ways to trade these things, most of them not working, building trading systems. But I had no clue what Bitcoin or Ethereum was. I just knew that market data was free, which was a huge relief. And like there were all these exchanges, were different prices on different exchanges.
Speaker: And then later on, like during the 2017 bubble, i ended up actually kind of engaging with some of the ideas, like reading Vitalik's blog posts. And I was like, oh, like there's actually kind of a whole intellectual scene here. Like a lot of it's pretty out there, but there are some interesting ideas.
Speaker: yeah But, you know, at first I really thought, had no clue what Bitcoin was. I thought maybe it was like PayPal or something, like some kind of digital wallet. did you Did you have any kind of a trading background or were just like, hey, this looks interesting. i want to try making money with this and then just see what I can do.
Speaker: It's funny. i So I had no trading background and I also wasn't interested in trading at all. Like finance, never something that interested me. i always liked computers. i like puzzles. I like games. yeah um But I wasn't even really sure that I wanted to work in in tech either.
Speaker: So I was really just just by nature not interested in any of this stuff. But someone who like an acquaintance who went to school with my older sister kind of convinced me that he had these great trading ideas and he just needed a software person to help him implement them. It's like he you know, he's smart. He's a math guy. He you went to MIT, but he was like, this this looks so easy. Like we we could do this for sure. I just need someone to help me write the code.
Speaker: um So in a way, I was lucky because that that would never happen today. Right. You you have LLMs. Yeah. And then ah most of the trading ideas didn't pan out, but we we had a lot of fun. and then eventually found profitable trading strategies.
Speaker: Was it difficult for you to to kind of grasp ah what trading is or how it works from just like getting his ideas and then trying to implement them? I mean, you you kind of get ah ideas already, but you still have to kind of um understand all the microstructure and all that stuff.
Speaker: Yes. No, it was extremely difficult. And I was also maybe it was this way for you, too. But like I was trying to consume as much content as I could. And I was so, so skeptical. I was like, this just sounds like a scam. Like this guy is telling me there's some like Fibonacci pattern or something, which to be clear, is a scam. It's just that the the information environment is so adverse.
Speaker: ah Like when when you started trading, did you run into any of this stuff? Was there something that didn't work that you tried? mean, I guess most of the stuff, because I came back in 2021 and then i followed all of the big accounts on CT. And obviously 2021 was like a very special time because everyone was just euphoric and bullish and 100k Bitcoin. And I didn't really like, I came from the Reddit part of of everything. That's where I first learned about stuff. And on the cryptocurrency subreddits, they were mostly like assigning random narratives to why prices moved and were like, hey, Solana got legalized in China or whatever today, so that's why it's up. And then when I came to Twitter, I kind of started realizing, okay, there's like perps and open interest and buyers and sellers and whatever. It's not just all tied to narratives. And then them I think slowly over time, I kind of like understood more and more, hey, um okay, this is kind of stuff that seems to work a bit more, just like intuitively makes a bit more sense. And like this person is just screaming for the price to go up or like this to happen, and they don't really understand right what a market really is or why things move the way they do. Yeah. Yeah.
Speaker: I think for me, like you mentioned, microstructure getting into like starting with this information environment of like, OK, how does trading work? Like, how does automated trading work? um i I think I would have had a really hard time if I just continued doing that.
Speaker: But luckily, like we started, we tried a couple of things kind of like really naive strategies to give an example. um I think a common strategy that you'll see if you start looking at like technical analysis stuff is an exponential moving average crossover where it's like you have a fast EMA of the price, you have a slow EMA of the price and you're like, OK, every time the fast EMA crosses the slow EMA, it means I should buy and vice versa. It means I should sell um to the extent that there that there is there are trends and and momentum in prices. It's not it's not a ridiculous idea. It's just.
Speaker: very hard to monetize and very hard not to fool yourself. I think if you're starting out with not a lot of ah statistical rigor, try to figure out like, OK, I back tested the strategy.
Speaker: Who knows how accurate my you know simulated transaction costs are? um And it seems like it works like 10 out of 10 times in the last week. Like, that's great. That has to be really statistically certain, right? yeah And you're like, okay, are these observations actually independent? So I think someone smarter than me could start from there and like end up at like a good rigorous statistical level.
Speaker: ah trading strategy. But luckily for me, like that was just clearly to work for me. I was absolutely sure that I was fooling myself. And so we went from there very quickly to um more high frequency strategies. And by high frequency, i don't mean that we were going particularly fast, more that we were looking at individual ticks and like trying to condition our strategy on individual updates on the exchange feed.
Speaker: And I think, you know, this is a good fit for someone who's technical because you're like, I don't understand this. I just trading stuff like let's just focus on like the WebSocket messages that are getting sent to me and how my program should react to them.
Speaker: Yeah. And honestly, like I would I think it's like a topical discussion right now, like if you should, you know, recommend that someone try to learn trading themselves or if it's even possible for someone to learn trading themselves.
Speaker: I i I think it's a pretty fraught terrain, to be honest, but I would recommend that anyone who's considering doing it, like actually try and look at market data tick by tick, not because high frequency trading is the only kind of trading, but because it's so concrete, like you won't lose yourself in these kind of like hand wavy. Like what what does it mean for there to be liquidity or like book depth? They're like, no, you're literally looking at the message.
Speaker: And then you have this tremendously challenging task of like, how do I go from looking at this message to writing a program that conditions on the message and like does the right thing? yeah And so, yeah, that that was kind of my that was kind of my introduction, like basically failing at the mid-frequency stuff, just being so overwhelmed and confused by it, and then going down to the to the basics at microstructure level. So that is mostly like market making or directional stuff as well?
Speaker: We went so we went straight to high frequency market making again, high frequency being like looking at every tick, though. I i think our system at that time, we had this terrible we have this cursed architecture. um I think our our internal tick to trade was probably like 10 or 15 milliseconds. So like pretty slow at HFT timescales and that not because we were doing anything interesting. We we were just doing it wrong.
Speaker: But yes, we were mostly doing different kinds of market making strategies, which also didn't work to be clear. We're like, all right, we placed bid, we placed an ask, like we're going to earn the we're going to earn the difference between these two things. And they're like immediately you're just losing money. um And actually, the first challenge for us was getting a high enough fill rate to make it obvious that we were losing money because we didn't have a particularly good research environment.
Speaker: yeah So it was really important that our strategy be trading often enough that we could actually just see like, yes, very clearly this was the wrong thing to do, even though it seemed nice in our research environment um and we moved from there i think like a lot of teams at the time like this was mid 2000 this was late 2015 right mid to late 2015 so markets are super inefficient we're like trying to make money quoting on one exchange and we're like hey what if we quoted on two exchanges like we'll have more order flow and maybe this will be better and then we very very quickly realized that the prices were just different on these different exchanges and like everyone else we were like oh okay arbitrage is the thing to do and then kind of got more sophisticated from there
Speaker: So you tried all of these other other things at first, kind of, and then just like went into arbitrage because saw this is basically an easier way to make money where you don't have to deal with all of this other stuff and kind of just like arbitrages. I think most people kind of like started, or a lot of people in back in those days had like did more quantitative or um automatic stuff and just ended up doing arbitrage a lot because that was the most profitable thing to do.
Speaker: Yes. Yeah, definitely. And if you see if you see a clear arb, I think um probably the first thing you should be thinking about is some kind of risk risk premium. But I'll i'll give us a little bit of credit like we we did consider this. I remember writing this little ah this little note on trading on BTCE which was this Eastern European exchange that eventually got shut down ah for for Russian money laundering. yeah But at the time it was like one of the respected crypto exchanges as respected as they get.
Speaker: And We had moved on from like this very mechanical arbitrage to like um what I would call kind of ah kind of a spread trade or basis trade, where like sometimes B2CE would be a lot more expensive than Kraken, and sometimes it would be only a little bit more expensive.
Speaker: And we couldn't actually withdraw from B2CE. We didn't have the banking rails on that side to complete the arbitrage. But we were like, oh, OK, we can buy on Kraken, sell on B2CE when it's expensive, and like do the inverse when the spread collapses. And we kind of modeled this out and we're like, wow, this is a really profitable strategy.
Speaker: And if we do it um in this kind of like high frequency style, we're also going to catch a lot of people who are just kind of like market ordering large amounts of money through the book. Like, this is great. They're like, wait a minute.
Speaker: This exchange seems pretty sketchy. So we tried to do some back of the envelope math on how likely it was that they would steal all our money, how much money we were going to make from the strategy yeah and then kind of like try and pencil out an EV from there.
Speaker: um And that sort of thing I think is really hard to get right because we assigned 15 percent chance annual that BTCE steals all our money. um And I believe we assigned a 5 percent chance to crack in.
Speaker: And I think in retrospect, ah like it did take about five years for BTCE to disappear. So I think in retrospect, we weren't we weren't too far off.
Speaker: But yeah you know probably it should have been like 30%. Yeah. yeah it's It sounds pretty pretty sketchy, especially back then. I feel like most exchanges, even like Even Kraken or like all the bigger exchanges, like nothing felt safe to the degree that it does now. like Nowadays, even after FTX and whatever, people are very comfortable with leaving their money on an exchange. Binance got hacked and then had their withdrawals disabled for quite a long time and no one really cares anymore. And back then it was way more of this, like not your keys, not your coins. like You need to self-custody or it's basically like, it's not your money at all.
Speaker: Yep. Yep. And, you know, these were like scrappy companies. I don't think anyone doubted Kraken's integrity, for example, but, you know, i don't know what their balance sheet was at the time, but I wouldn't have been surprised if it were like $30 million dollars or something.
Speaker: ah Whereas Binance can just eat 6 billion or 5 billion. No problem. Yeah. even Even Tether, I mean, I guess now we would probably never know ah the truth, but I think there's probably a case we made that there weren't always 100% solvent like the whole time throughout their history. Certainly. I mean, even if it like worked out now, but.
Speaker: i'm I'm a Tether truther, I'll say. It's my like embarrassing confession. I always have been. But as you say, they've made so much money in the inter intervening time. I don't think they're going bankrupt anytime soon.
Speaker: I mean, at this point, it's probably even even just like a state level thing that can't really happen because they hold like so much in treasuries or whatever. and And it's a very different environment than it was back then.
Speaker: But but like when you when you um you you did the the arbitrage boards and then you started being profitable with that, and then how did you progress from there? Did that like keep being a profitable thing or decrease over time? And did you move to other things then?
Speaker: Yeah. So we had, you know, we we had this kind of journey where we went from like knowing nothing to learning how to do arbitrage profitably. And I would say most of the most of our progression at that time was was on the engineering side. Like we figured out how to make a good trading system, make it replayable, have a simulator um and then some some amount of like selecting the right venues, figuring out how to have banking rails, all that stuff.
Speaker: And then eventually adapting strategies like I described, that spread trading strategy to work well without banking rails, doing triangular arbitrage, stuff like that. And that worked pretty well for us through like early 2017. And then this kind of like GarageBand two person operation I was part of, ah we the we were actually in Mexico City at the time that and we were sitting in the office of the Mexican exchange, which was tremendously cool because, ah you know, it's not often that you get to sit in the same office as the exchange and like talk to all these smart guys who are ah you know, building systems. There's there's this little crypto community.
Speaker: ah we're meeting local entrepreneurs like poker players and very like mid twenty ten s vibes. um And one of our main trades was basically the spread between this local exchange and Coinbase in the US. Like pretty, pretty classic cross-border ARB.
Speaker: I like to think we applied kind of like our own secret sauce on top and we were able to do things like the spread trading and develop some some ways of squeezing a few more basis points out even when we couldn't really send more than a certain amount of wire transfers per day and stuff like that. But in the end, it's like it it lived or died on the basis of this spread.
Speaker: And over time, people came in and competed away. Like our one of the guys who's who's still a good friend of mine today who really competed with the spread for us was not a particularly strong trader, but he was a tech entrepreneur who had all these companies.
Speaker: And because of that, he had these like 10 year banking relationships that were just rock solid. And he was able to move pesos to dollars back and forth very, very easily. So, you know, people like this started coming in long before professional firms. I think at this time, Jump was still using crypto for like their interns. They were like, you can go trade here for a little bit and then maybe figure out that like, we'll let you trade something real.
Speaker: And so so there weren't like these big sophisticated firms competing things away, but just people with these kind of ah like FX advantages, frankly, were able to compress the spread incredibly.
Speaker: So at a certain point, our trade went down and we kind of decided like, oh, we should call this off. I became interested in inc crypto a little bit, but I had also been bitten by the by the trading bug. So I started off like, I'm not interested in finance. I'm not interested trading. like That's like dumb and intellectually satisfying.
Speaker: And then at this point I was like, wow, this is really cool. It's like, you know, some kind of competitive programming scavenger hunt where like you get this crazy feedback loop and like most things don't work, but there are all these different ways you can look for edge.
Speaker: And I was trying to figure out like how I could learn more about this. And so I made a decision that was good for me, but was bad in context, which was I went to work at a bank. So I i went, I accepted a consulting arrangement at this bank in Liechtenstein.
Speaker: who was kind of in the process of transitioning from old school private wealth management to like new fintech friendly crypto first banking. So they were they were going to be like a good banking partner to crypto companies.
Speaker: And they wanted someone to help them figure out how to like set up a custody system, how they could actually access crypto markets to like buy the stuff on behalf of their clients. um And then also who their client should be, because this was this was 2017. So like people were coming out of the woodwork with their like white papers. I'm sure you remember some of these some of these scams where people would just like take a white paper and then take a generic image of like someone who looks like they have a Ph.D. and i like put it on the white paper. be like, please send me your Bitcoin. You'll make so much money. are a lot of scams.
Speaker: um And so I spent a few months in in Lichtstein helping them kind of like set up the system, learning about how things worked at the bank, like how their trading desk worked, which was nothing like it, like a HFT trading desk. Like these guys were mostly going out into the market and just like executing orders on behalf of their their rich clients. um But that was fantastic because I learned a lot more about ah First of all, my kind of like selfish interest was like, man, we've worked so hard to keep our bank accounts open. Like what's happening on the other side of the of the curtain when they're evaluating clients?
Speaker: And I also got to meet a lot of interesting crypto companies. Like I got them set up um along with the rest of the team. Honestly, like the the person who was leading this initiative at the bank was very forward thinking and he wasn't he wasn't a crypto native guy, but this whole thing was basically his idea. He was like, we need to to get someone to help us set up this crypto division. There there are all these good clients out there.
Speaker: So we, you know, we started banking Bitstamp, Kraken, company called Bitpanda, which is still around. You don't hear about them as much, but um they're They're like a pretty big European company and then Bitfinex for a time as well as well as plenty of other their clients.
Speaker: And so it was a cool way to get in touch with like what people were doing in the crypto ecosystem. And i also met a lot of colorful characters who were doing other other sorts of things and trying to come to the bank to open accounts for like their ICO proceeds. And I think we we probably filtered out a a decent chunk of of those folks.
Speaker: um But eventually it was kind of like, OK, I understand what's happening here at the bank. It's um it's cool. They are building, frankly, an extremely successful business, but it's not it's not trading. It's not that scavenger hunt that that I really like. So and it also felt like
Speaker: Like i was meeting the people that I thought would be the most sophisticated players in these markets. And that was helpful for me to understand actually just how immature they really were. Like, you know, i was very cognizant of the fact that our HFT system was not the same thing that would be running at Jump.
Speaker: Right? Yeah. um What I was not cognizant of was like, I guess the opportunity costs for these large firms. Like in in my head, it's like everything's going to be perfectly efficient really soon. Like it's not worth investing in this stuff.
Speaker: But I sort of came to realize the decision making processes that must have been happening at these places where it's like, yeah, are we going to like spin up this whole operation, take some crazy compliance risks? because Now I had some experience dealing with a compliance department ah to earn like two million dollars a year on some trade. Like absolutely not.
Speaker: And I'm sitting thinking you like, I would like to earn $2 million dollars a year on some trade. And these guys aren't going to do it. So, so yeah, I went back to Mexico and I started along with two the people that I had met when I was there, a little crypto market making firm.
Speaker: And we felt like, ah We kind of had some idea of how to build a more professional system. We thought there was room for this in crypto. And then also ah with the advent of derivatives on BitMEX, we were like, clearly these markets are going to grow in volume quite a bit.
Speaker: And we were, you know, we we were excited to to compete for a chunk of that. Mm hmm. And over time, obviously, like, you know, we set up this sweet trading system with all the lessons learned from like the first couple iterations that I had done.
Speaker: And we started out with some like pretty, pretty basic strategies, like mostly trading spot. um Then we started trading features on BitMEX.
Speaker: I even emailed Arthur Hayes at some point being like, hey man, like you read all these blog posts about how the, you know, how the futures basis is supposed to work as a result of, you know, all these macroeconomic analyses that you do.
Speaker: Like, how would you think about combining those and kind of like a quantitative strategy? He had also done the seminar video on like, here's like a little Python market making bot. And I was trying to square the the, like, how does this like big picture macro stuff like actually play into this? um Amazing guy responded like basically immediately. ah He didn't literally say none of this macro stuff matters, but he didn't mention it at all. He was like, here's a handbook on derivatives modeling. Like, here's how you should think about the future's basis.
Speaker: yeah And I was like, man, OK, thank you so much. Also, it's interesting kind of like. I was like, man, I've been reading all this guy's blog posts and like half the stuff doesn't make any sense to me. And it turns out I just don't even need to think about that at all. Like that's so not relevant for anything that I'm doing.
Speaker: Um, so yeah, I just went from there and we ended up running this, this company is like a five person trading team for, uh, for about five years, like late 2017 through end 2022.
Speaker: it's It's kind of funny you say that about Arthur, because this was like back in the day, but I feel like it kind of still applies her to what he's putting out ah in terms of content nowadays. Because I kind of like read a lot of his ah his blog posts, because they are they're entertaining, but like over time, i think like two days ago or something, someone made a post saying he's just a like inventing ways to to be like, hey, they're going to print money. And I feel like not it it like it it it does kind of make sense, the stuff that he's like outlining, but it's just like another way of like, hey, this is how they're going to print. or like And then once in a while, it's going to be like, OK, but but also buy this shitcoin, like this this ah right special whatever that that that I'm in because it's going to do this. And yeah, it's a lot macro LARPing, I guess. but It's it a lot of discretionary macro.
Speaker: Yeah, which um very very interesting interesting to read. Yeah. And this, like speaking of adverse information environment for for people trying to learn trading, like this is the number one thing I would say not to pay attention to. Because how long is it going to take for you to learn whether you're doing something right or wrong if the thesis is like in five years the dollar is going to hyperinflate so you should hold Bitcoin that whole time?
Speaker: Like what's your feedback loop if you're doing that kind of trade? without really strong priors, like working at a firm that can can do 10,000 of these trades at the same time. So I would advise anyone to really just stay away.
Speaker: Especially because if any of these things actually happen, then you probably have a lot of other IRL actual problems instead of just being like, hey, what do I do with with my internet money or whatever? But it' it's funny to contrast because yeah as you mentioned, also the this this market making series that he made or like the videos on the BitMEX channel, I think those are actually yeah very good substantial actual info that is like still out there and a good place to start for people.
Speaker: I totally agree. And I think there's a hugely underrated element in that code base also, which is that um most exchanges which write an SDK and kind of show you, you know, how you could run a basic market making strategy, have this structure where they're like, OK, we're going to have like we're going to listen to WebSocket connection and then we're going to have this event loop where we're like pulling the exchange and then inside the event loop, we're going to like go off and place orders and maybe we're going to use like the languages async feature. But it's like everything is happening in this one spot. It's because it's an example and that's how you that's how you understand what's happening.
Speaker: But um if you build a trading system this way, first of all, it's going to be very hard to have discipline about actually like managing your orders and understanding where they are in the book.
Speaker: because you need to be able to make decisions at all times based on knowing that you have a pending order that's about to enter at a certain level while your estimate of fair price is updating and that deciding what you're going to do on that basis.
Speaker: Whereas if you're if you're managing all this stuff inside this one little loop, like maybe your code isn't even called until your order finishes placing. So it's just absolutely useless as a basis for someone who's then thinking about ways to improve the system.
Speaker: ah One thing that BitMEX did really well was they didn't do this. They had like one set of orders that the strategy computed that was like, here are the orders that I think I want open. And they had this separate function that they called converge, which was like this monster function. It was not very pretty. But the fact that they separated it out was fantastic because the separate function would look at those orders, look what you actually had to open and then figure out like the series of places, cancel, amend that you would have to do.
Speaker: um So I thought that was fantastic.
Speaker: So how how did your your ah market making operation continue to develop and why did it come to an end in 2022 besides ah maybe obvious reason? Yeah, i think you probably guessed why we decided to wind down.
Speaker: But um I'll leave that for the for the end. I think, how did we develop? Like we like i said, we started trading derivatives on BitMEX. So we started with the calendar features and then ended up trading the perpetual swaps as well.
Speaker: At a certain point, Binance futures sprang up. We started trading there. were trading on Bitstamp. Like we probably integrated like 15 exchanges over our lifetime. And then at any given time, we really may be like seriously trading on three.
Speaker: So we kind of like go through these um these cycles of like building a strategy, deploying it, it would be going well. We'd be like trying to build new strategies in the background. The alpha would decay and we we'd have to switch over to a different strategy.
Speaker: and We kind of skipped the era. There's a while where exchanges were adding margin everywhere. It was like kind of this weird space between spot and perps where they were like trying to replicate, I think what they saw working successfully in the stock market.
Speaker: it was just so painful to deal with, like from a from a trading system point of view. And then just like, what's the bar rate going to be on this margin? And um I'm sure there was a good way to do it. and I honestly haven't given it any more thoughts since then, but we skipped that entirely and went straight to perps.
Speaker: which have a lot of cool properties like ah the funding rate obviously is implicit in the in the price of the perp. um I think if you're if you're the kind of person who is looking for technical edges, it's really nice that all the exchanges have their own quirks about like how they how they implement their funding rate calculation and um what kind of like index they use to mark the perps and and all that good stuff.
Speaker: ah So we ended up trading a lot of perps and then, you know, we were in Mexico City this whole time. So we were also trading on the local exchange, that local Mexican exchange. But so then kind of expanded through the rest of Latin America. So we were doing some some market making for them in Brazil, Argentina, Colombia as well.
Speaker: Often just finding ways to provide liquidity for people who had the local currency and wanted to get out of it into Bitcoin or dollars, basically. So we were finding ways to Like one one really naive way to do this without any banking rails that was surprisingly useful at the time. I don't know if you could still run a strategy like this is just figuring out a way to look at, you know, say the exchange has three pairs that actually get volume like BTC, ETH, and at the time, probably XRP would have been the third one. Just figuring out how to take those three order books, um use your fair price and your order flow that you have in those three assets in dollars.
Speaker: and then use those three assets on the exchange in its local country and its local currency to kind of aggregate liquidity across the three. You can kind of take, say, like a BTC Mexican peso order book, combine it with a BTC USD order book to get an implied USD MXN order book. You could do the same thing on Ethereum, same thing on XRP, etc. So if you have a system that can kind of do all of that, aggregate all that together, you're not doing anything particularly ah You know, it's it's not rocket science, but you're kind of just like putting together a little product that is useful for whoever is trying to go in and like market order BTC so that they can get out of their home currency. Mm
Speaker: hmm. So so, yeah, we you know, we we expanded in those ways. And then 2020, in twenty twenty This was a big mistake. We missed MEB completely. Like, I remember it was happening.
Speaker: Fun fact, I was not on Twitter this entire time. I'm not sure if it would have gone better or worse for me had I been. But I had to had very limited understanding of what was happening on crypto Twitter. My my business partner, Josu, he was all over Twitter. So he'd come in with, you know, i'd I've learned the important things maybe 12 hours after they happened or whatever, but I was not on Twitter at all.
Speaker: And so we were we were looking at this MEV stuff and we were like, man, this is not going to last more than like a month. Like no one's going to keep trading on a protocol where you can just be front run all the time. um So like, yeah, we could probably be good at this, but like not not a good, not a good use of our time, which was obviously a terrible, a terrible take to have.
Speaker: But nevertheless, like we we kind of stayed out of DeFi until DYDX v3, which was not really DeFi, but it was ah basically the thing that drew us in was that the API was basically like a centralized exchange um and they had this cool, you know, zero knowledge proof to settle back to Ethereum or whatever. But for us, it was kind of like, oh, cool, a new perps exchange.
Speaker: And we'd been introduced to them by by someone that we we thought was credible, basically. And we're like, oh, this is cool. It's like it's like a purpose exchange, but we don't have to like KYC and open a bank account. um So in 2021, we were, i think, the very first market maker to be integrated on DYDX v3. I know Wintermute was like working on their integration at the time, but I think we actually finished it first and we were like live trading, I think the very first liquidity provider on DYDX v3.
Speaker: And so for the first few months, like before they launched the token and whatnot, we had pretty decent market share there. Like there was a month where we were 50% and then you you know kind of slowly competed down as other people entered the market.
Speaker: um That was a lot of fun, honestly. Like, you know, you mentioned 2021 was like this crazy year. So many things happened like in the lead up even early 2022 with like that Terra Luna crash.
Speaker: um Yeah, there's just a lot of stuff happening. And if you were if you were a team that was like looking for mechanical edges, you could spin up on a new exchange quickly. ah There was a lot that you could do, even like chain forks, like every time there's a chain fork or like an airdrop, there was something to be done. And then, of course, FTX was like huge at this time, like really growing in stature, tons of volume, all sorts of interesting things you could trade. um so we were having a ball like 2021 was great year.
Speaker: There was like new and interesting strategies everywhere. 2022 was like stressful, but good. Like there's all these trading opportunities. And then know I remember going to the FTX Bahamas conference and going to Bitcoin Miami, which is like not my environment at all. Like I'm not, I think by nature, I'm not by nature like comfortable.
Speaker: don't fit in with like the Bitcoin Miami community, let's say. Yeah. And then also just like conferences, not really for me. But I remember going to these two events and being like, wow, like this is getting this is getting crazy. Like maybe we should be maybe we should be short. Like we never take discretionary positions. Like I think that stuff is not something I have any skill in.
Speaker: But I remember getting at this point be like, should we just buy some put options? Like I hear Jane Street buys put options sometimes because, you know, they're their their P&L is really correlated with um with stock market volume and you know when and when it crashes the volume goes down I had this like a idea I was like maybe we should do something like that because like it seems like this thing's getting pretty frothy and like if people just stop trading all this stuff then you know our P&L is going to go down even if we're not taking market risk um so we did a little bit of that and then FTX happened and we were
Speaker: I think, you know, I mentioned that early on for that like BTCE crack in trade, I tried to like sit down and do this, do this estimate of, you know, likelihood of them going off with all our money.
Speaker: By the time, like after two years of complete ah you know, the whole world's upside down in this in this like weird crypto environment. I think some of that some of that natural suspicion had gone out the window for me, and we had like a third of our book parked on FTX. yeah But honestly, like we we tried to do this risk assessment with DYDX because we were like, listen, this is some like weird zero-knowledge proof smart contract thing. There's a pretty solid chance that something goes wrong we just lose all our money. Yeah.
Speaker: And so the the irony is we were so careful about the funds that we kept on DYDX and like, you know, we had all these monitors and whatever. we were like, sure, yeah, FTX is definitely way safer. Also Binance, like we had so many we had so many cases that made us mistrust Binance in one way or another.
Speaker: um Also, their involvement with Tether, like were like, yeah, like Binance is pretty sketchy. You know, we we definitely don't want to have more than more than 20% on Binance. We don't want to have more than 10% on DYDX FTX. It's like, yeah, these guys, it seems fine, like 30% there.
Speaker: ah You know, wrong call in hindsight. But um Yeah, so so we ended up taking like a decent loss in FTX and we were thinking about whether we wanted to keep going.
Speaker: um And at the time it was like, it's hard it's hard to remember this now because you think of FTX as like this like this deep crypto winter afterwards. but ah By this time, like Jump, Tower, Jane Street, like every really good trading firm is trading crypto.
Speaker: And we're like competing with them, right? We're we're like, oh, okay, I guess we can't be top of book on like BTC, USD, on Binance anymore. Like maybe we can go trade like MANA and whatever, or like maybe we can go to the, you know, this exchange in Argentina and we'll have really good market share there. So, you know, we're definitely feeling the competitive pressure ah Even Wintermute, like Wintermute was pretty small in 2021 when we were trading kind of neck and neck with them on DYDX. And both of us got obliterated by Tower because they had physical infrastructure that that both our teams lacked to connect Binance to the DYDX matching engine. And we were just totally, totally caught off guard by this.
Speaker: And I think so, we you know, we were kind of starting to perceive that the space was getting more competitive and then You know, at the time we were talking internally before FTX happened, we were like kind of plotting on a way that we could hire someone from one of these firms because it was clear to us that we needed, um you know, there were things that we were figuring out on our own that we were really proud of having figured out.
Speaker: But it's like, man, if someone had been at one of these firms and like just really knew what we should be doing, even just describing the objective function to us that that we should be looking at here, um it would have saved us so much time.
Speaker: ah So we were already thinking about that. And then FTX happens. It's like, all right, I've i've had enough like this. It's so fun running a five person training team. But I also felt like um it would be it would be cool to work in a larger organization with, you know, a ton of really smart people from different backgrounds that I can learn from.
Speaker: So so yeah, after FTX, actually, i I took some time off, but I ended up going to work at DYDX because I'd gotten close to them over this time or are moderately close. And I thought they were kind of some of the highest integrity, smartest people that I'd met in crypto at that point.
Speaker: um What were like the the things that over all of this time that you that you ah did that were the most profitable and what were some things that you spent a lot of time and effort on that turned out to be like kind of useless in hindsight or like comparatively?
Speaker: Yeah, that's a great question. I would say most most profitable as a category was just we had this we had two things that were kind of scrappy. We had like this really kind of dirty data collection system that we didn't we didn't use for for everything, but like we could add a new exchange to it in like an hour and we'd have like OKish data.
Speaker: um And then we'd have I guess this is no longer impressive, right? LMS can do this immediately. But at the time it was like, all right, we you know, that that's a that's an achievement. yeah um And then we'd be able to mark basically every trade that we were seeing through this data collection system against our idea of fair value at the time.
Speaker: So it was really easy for us to look at a new instrument or an exchange, capture everything that's on the exchange and be like, oh, interesting, like for the last week, there has definitely been an opportunity if you're if you're a market maker on this redacted token on this exchange and similar for for funding rates.
Speaker: um So just a having the ability to like quickly notice that something might be interesting and then B, we had this one really adaptable strategy that was kind of like a spread trading strategy with a very um very permissive inventory skew layered on top.
Speaker: So every time we would deploy the strategy somewhere new, it would start in the same way. It would basically, there'd be two instruments. It would quote on one instrument as a maker at the price where it could hedge on the other instrument as a taker, and then just add some edge to that initial quote.
Speaker: That's how it would start. And then it would skew its quotes that on the side where it's where it's getting aggressed and taking on kind of inventory with respect to the spread. um It would skew like proportional to X over one minus X. So it just kind of like goes up towards infinity as you get filled more and more.
Speaker: And this is useful because you have some starting edge. You maybe haven't done a lot of work to figure out like actually how big the spread should be between these two things. um But you have this inventory skew, like the market will kind of teach you like how how close the spread should be. And you know, your inventory will hover somewhere.
Speaker: And then the other side where you're not getting aggressed skews linearly. So it's never it's never um like in the worst case, it sets your edge to zero. And this is like a kind of a terrible trading strategy if you think about it. It's like you have you have no real edge like you're not bringing information to the market.
Speaker: And yet if you just deploy this somewhere very, very quickly, you now have a set of um your own fills you have market data and you can start layering stuff on top so you could get set up trading something like same day basically and uh doing this kind of like spread trade like can even be profitable if you have kind of like a chart like a dashboard that i described and you're able to kind of eyeball like hey i think the spread should be like somewhere between 20 and 80 basis points you're like great okay i'm gonna set this up um You're getting fills when people like market order through a thin book. So like you're kind of stink bidding with respect to the spread um and you don't take a ton of inventory risk. So like I would say that strategy by itself was the single best kind of effort to profit ratio that we had because you could just get set up, start making money and then start kind of layering stuff on top to figure out like, OK, what should the actual fair value of the spread be?
Speaker: ah um And then, okay, so you asked two things, like what what didn't work? There's so much stuff that didn't work. I think I spent a lot of time implementing papers early on. And I won't say that this whole process didn't work because I think it's a valuable way to gain intuition about what does and doesn't work.
Speaker: ah But yeah, like most the papers didn't work out. I remember working on this thing called um ah published by a guy named Sanme Das called a Bayesian market maker. And I was attracted to it because I was like, wow, this is like a Bayesian model for figuring out what what fair price should be given trades. And it's so, so different from everything else that I've seen about like classic high frequency signals like order book pressure, trade flow and balance, order flow and balance. Like this thing's just totally different. Like it would be so cool to build this model.
Speaker: and I got kind of nerd sniped. I think I spent, embarrassed say, three weeks on this on this paper, which was like not even supposed to be that good. It was like this guy's master's thesis. There was no reason to expect it it would really work.
Speaker: um But, you know, I was convinced. I was like, okay, I'll take this basic thing and I'm sure I'll i'll have some kind of like Eureka improvement on top of his model that that will make this thing work or I'll find a way to mix it with our things. And it was just a total waste of time, I think.
Speaker: But it was still a lot of fun. Mm-hmm. Did your ideas end up learning something from it or like? Yeah, I think so. um I think so. I mean, it it pretty quickly became clear that
Speaker: you would you would just end up if you're trying to apply a pure Bayesian approach to something like the fair value of an asset that's moving over time, at least the way that he implemented it, if you kind of assume that you're going to be able to update towards some ground truth, that's just like clearly wrong.
Speaker: It's almost a category error because your fair price is moving over time. So like a signal that you get. further forward in time ah maybe is not really comparable to a trade that you saw 10 minutes earlier that was telling you about the fair price 10 minutes earlier, unless you already assume that there's some predictive relationship and.
Speaker: Yeah, so as a result, the whole thing that this is why it was a waste of time, right? I should have looked at this and be like, all right, this is like weird. It doesn't totally make sense. yeah But instead, I was like, oh, I'm sure there's like some way that I can go in and like correct for this and and turn this into something useful.
Speaker: ah And yeah, so it it was it was humbling. I'll say that. I see. So um before we we move on to what you're doing now, what what did you do at DYDX and what was your time like there?
Speaker: Yeah, do do idx so my job at DYDX was I led the research team. And research is like a pretty poorly defined role at a lot of these DeFi protocols, I think. But the thinking was basically like they were trying to design a totally decentralized version of their V3, which had a centralized matching engine.
Speaker: And... When you decentralize the matching engine, you know, performance gets worse, all this stuff. But also ah you open the door to ways that trading firms might kind of game your system in order to get better P&L, including but not limited to m MEV.
Speaker: And then this has some effects on kind of just your ability to have like a useful market and a good product. So the thinking was like, all right, i used to work in trading firm. I was always trying to reverse engineer exchanges and figure out ways to like you know, take advantage of weird ways that they had heads up their APIs.
Speaker: Like maybe I can be useful as a second pair of eyes on this design for for v four um I think that was kind of the initial idea. And then going in, it was like before was already basically completely designed by by the time I joined.
Speaker: So then it became like um a combination of basically like figuring out how to have a good offering for for trading firms, not just for like prosumer retail traders. And then also, um you know, like how should the insurance fund work? Like how is ADL going to work? How is how is liquidation going to work? And like, how can we do this on a blockchain?
Speaker: And it was super interesting because Uh, you know, I said the thing I wanted was to work with a bunch of talented people and I absolutely got that. The the people at DRDX were fantastic. Nevertheless, we were, had a front row seat to us, like getting our asses kicked by hyper liquid. Like it was incredible. The degree to which they were beating us as a 10 person team. Yeah. We, you know, we're like 50 people. we have all these, you know, I had all these colleagues who are incredibly smart, uh, from places like Google, you know, crazy academic backgrounds.
Speaker: Um, we had balance sheet with hundreds of millions of dollars and Hyperlux was just kicking our ass. And this is clear to us before, I think, most people would expect. Like, I think if you looked at the situation from the outside, you'd be like, oh, like, they're probably kind of complacent. And ah certainly we were, but it's, I think from...
Speaker: I first heard about Hyperliquid in May 2023, I think, on Corey Hofstein's podcast. remember Jeff went on and and did an episode. I was like, oh, this exchange sounds cool. And then internally in Slack, by the end of that year, it was like all we talked about DYDX. It's like, man, these Hyperliquid guys are really good.
Speaker: Like this is before the airdrop. And um yeah, i I mean, obviously learned a ton from that, but I think more than anything else, it was just tremendously inspiring to see that
Speaker: you know I'd come in kind of prepared to accept this bitter lesson that like the better product doesn't always win, like it doesn't really matter that you care to do you know such such a ah craftful approach to your job. like There are real economies of scale and like you want to be at a well-funded company and just building a better product is not enough to to win. And i I still think, unfortunately, like that's that's true in some ways, like we want it to be one way, but it's another way.
Speaker: However, Hyperliquid really showed that like a 10 person trading native team was able to just challenge like a well-funded VC backed tech startup and like totally eat their lunch in a very competitive market with certainly hundreds of millions of dollars, maybe maybe more of revenue on the table.
Speaker: um And so maybe that's a good segue, right? That that was really inspiring for me and my co-founder, Ivo, who was the CEO of DYDX for a time when we decided to start Pascal. So we've seen this happen with exchanges over and over again, right? It's like if you're like actually just hungry, you really care about building good exchange. You have a small team that's able to to hold a lot of context about how people are actually going to use the exchange and you're able to go quickly. There's a lot you can do.
Speaker: What do you think was like the the the main thing that um prevented or that that made DYDX fail against Cyberliquid because it it it did have this good position, as you said. But I do remember that for me personally, just from like a user perspective, I also noticed Cyberliquid more at the end of 2023. And it wasn't like perfect or great or anything yet. DYDX was cool, but it was never really, I think the user experience was kind of like lacking a bit. do Would you say that is what made Hyperliquid win at the end?
Speaker: Yes, and i maybe I would go a level deeper and say, like, why is the user experience difference different? Certainly there are a lot of technical choices, but I think what it comes down to is ah The Hyperliquid team, like those guys were traders and they were trading on their own exchange. Like they had to operate hlp So, you know, if the API told them that their order was canceled and later on it was like, hey, by the way, you filled, they would be upset about that. They'd be like, what the fuck? Like we have risk that the exchange lied to us about what is going on.
Speaker: Whereas, you know, particularly if you're not allowed from a compliance perspective to trade on your own exchange and you're so large that most of the people you've hired are not really, you know, maybe they don't even know what an exchange is.
Speaker: uh it's really really hard to compete on on user experience
Speaker: i see it and also it makes a lot more sense now that you said it um why you decided to build what you're building now which is a prediction market if if that is the correct term to use and it it's very very interesting i didn't think of it that way before because um there are like these these huge players already in in polymarket and culture And um from Kachi's side, I don't know too much, to be honest, about like their their architecture whatever. But I've read a lot of times that Polymarket's API is like very... Or like not even just their API, but their whole like system they're matching and whatever is very lackluster, let's say. So it definitely does make sense that you're building something that tries to... like
Speaker: fill a niche that is needed there. um Maybe you can describe what what Pascal is and what what their vision is for it. yeah Yeah, absolutely. Can I ask first where, um I always like to ask people this question, where you first heard about prediction markets, like what context it was in and and what you thought of them?
Speaker: I think it must have been Polymarket and it's been quite a while. It was during like... I feel like in 2023 because they they have been around for quite a while and it was like during kind of around Frentech like when all of that happened in like the 2023 bear market when they were still like the only thing to do on on Polygon. Yeah.
Speaker: And I guess I never really thought about a prediction market is kind of a concept that just makes sense. like I didn't really i didn't know what it was before, but it's like not difficult to grasp. I think once you see what it is, it it's just like the the thing that you maybe know from like sports betting or whatever already. And then I was like, oh, cool, I can do that on chain. And yeah, that's how I got it. Yep.
Speaker: that That makes perfect sense. For me, i think it it was less intuitive for me because the first time I heard about it, I think was in an essay that Vitalik wrote. so this was probably 2017. And so for me, it was very in this milieu of like these crazy utopian crypto ideas where I think he was referencing Robin Hanson's Pewtarkey paper. And he was like, he went very quickly from like, you could have a smart contract that settles based on this outcome to like, here's how you could reshape the whole government to make policy decisions that way. And I was like, yeah what? Like, this is crazy.
Speaker: ah Um, so I kind of put it in the same bucket as like, I don't know, DAOs or some of these other like really out there crypto ideas that were kind of almost utopian in nature and kind of like provocative and like sure, pretty unlikely to work out. But there there's something interesting there.
Speaker: And then. I remember trying Augur and that it it was really challenging. Like they had this, it was it was a chain, right? And the way that they settled markets, like the entire chain would fork if the resolution wasn't agreed upon by all the stakers. like it was a very crypto mechanism.
Speaker: And so that the user experience was not quite there. And I almost didn't even think about it as a marketplace in the same way that I i thought about. Like I was trading perps professionally at this time, ah and i I almost didn't even think about Augur as like an an order book that I could trade in a high frequency way.
Speaker: um And then i remember seeing Polymarket probably 2022, late 2022, still during COVID for sure. And they had these markets on case counts, like COVID case counts.
Speaker: And they also had markets on policy decisions that were going to happen in, I think, California and Texas or maybe California and Florida. So there was a very like red versus blue coded response to the pandemic. And then there was actually like, ah like something you could predict, which was the case counts.
Speaker: And I had this flashback to that Vitalik essay and I was like, oh my God, like, you know, sure, there's not a ton of volume here, but like someone's actually like kind of trying to make this work. Like you could you can come in here, you can kind of like pencil out exposures and how these things move together.
Speaker: um and if you think you have an idea about what this policy is going to do to case counts in a particular state, like you can actually trade on that basis. And I think I told you I thought I thought, you know i was never really interested in finance or like crypto underlies at first.
Speaker: And so for me, this was like the idea that there was actually some fundamentals underneath the thing you were trading was like crazy, mind blowing. Because to me, trading was like that, like microstructure kind of like jigsaw puzzle, which is super fun. But like at the end of the day, who who really cares about leverage? Dogecoin futures or whatever. And I saw prediction markets. i was like, wow, this is so cool.
Speaker: And like maybe there is actually a way to build to build a full marketplace around this. So I remember playing around in Polymarket and I was kind of like back to my typical style of trading where I'm doing these very mechanical things of like, oh, um I guess it turns out that before holidays, like holidays make the case counts go up.
Speaker: But they also make people not report the cases because they're not working on the holidays. So there's this kind of like mechanical trade you can do around the holidays because most people are pricing based off the the published case count numbers. And, you that was a little bit like realizing that there's a derivative that's marked to an underlying index that has some quirk about it. And and you could trade that.
Speaker: But at the time, they were still on an AMM, which seemed crazy to me because it was like these things resolve to zero or one. So like clearly you will lose all your money unless you pull it out of the AMM right on time. So I didn't really understand You know, I had a full time job. i was kind of like, this is a cool thing to do in my free time. I didn't really understand like what what is happening here and and what's their plan?
Speaker: And sure enough, they got rid of the AMMs, they added order books like that was awesome. And I think over the intervening years, like they really popularized this product like it was huge during the 2024 election.
Speaker: um And I remember watching that and thinking like, wow, this is so cool. But also I had kind of just assumed when I saw this in 2022, that by 2024 or 2025, there'd be like a totally different exchange where these things would be quoted.
Speaker: Like they'd probably be quoted on when I first saw I was like, I'm sure these will be quoted on FTX. And then I was like, oh, probably like on Binance, like you're going to have this and like um you're going to be able to trade like really tightly. Like what's, you know, how does the presidential election, know,
Speaker: affect Bitcoin prices. And like, there's obviously going to be a tight relationship between this one instrument and another. um And I think we basically still don't see that. Like, this is not necessarily the exchange's fault. Right. But I think if you look at returns in like who's going to win the the Senate for midterms, like you look at returns in that time series and then you you look at returns and in Bitcoin price, ah actually not so related, like no one's no one's really making that thesis and and doing a trade on that basis.
Speaker: So. I think there's still I think there's still a long way to go, but I was just really captivated by the idea that that you could have something that's like has all these elements that I like about trading, but there's some real underlying thing that's attached to it, which was like so obviously missing in in a lot of the crypto stuff that I was trading. Or at least yeah maybe maybe there are people who are capable of making fundamental assertions about what will drive Bitcoin price, but I am not one of them.
Speaker: Right. Like all of those things are like gibberish to me. um Whereas if you're like out trying to predict COVID cases like that's a tractable thing. um So that was kind of the context that that got me excited about prediction markets is like, wow, there are all these people, these like small teams who are doing
Speaker: analogous things for elections that I would kind of think of as these like mechanical trading things that I would do in big Bitcoin. Like they they look very different because they're mostly not like high frequency nerds, but they're like out there doing like nerdy things and trying to collect data and like trying to be scrappy and find different ways to get edge to to predict these outcomes.
Speaker: And I think there's so much you can do on top of the existing exchange offering. So like I really respect the the existing teams for being willing to be the first ones through the door and like put something out there, especially at a time when it was less obvious that that was a good thing to do.
Speaker: I think right now it's really sad to see they're kind of jettisoning the international market. Like it's clear that just being a sports book is so profitable. And so there's a lot of focus from from both of the incumbents on like be a sports book in the U.S.
Speaker: forget 10 like parlays, 200 like parlays, like make this easier, lower the friction. Yeah. um Which, you know, would be unrecognizable from like a Vitalik essays point of view. And on the other hand, in these international markets, like you're seeing these really cool things like we have We have global conflicts that are happening, especially in the past couple of years that people are actually investigating and bringing money to the market, bringing information to the market, excuse me, through Polymarket International.
Speaker: Like that's so cool that that is happening. And if you if you go back and look at at like a strike ah like U.S. striking Iran, you can actually c like you can plot the time series of the relevant contracts on Polymarket versus like crude oil on hyperliquid versus crude oil on CME. And you can see like when the information leaks and how it diffuses through these different markets, which which is incredible to me.
Speaker: um so there's so much benefit to be had by ah just having an international product with like interesting real world underlyings. And then I think making a defy just makes sense. Like you it's fundamentally inherently an international product. Like there's no there's no reason it should be us only.
Speaker: i think lowering the friction to the point where you're like, you know, doing it from your phone, it's really easy to onboard is like cool, but it's it's also it kind of selects against the archetype of person who's going to be like bringing useful information to this market in the first place.
Speaker: So I just think there's a lot to do yeah on the DeFi side. Like, um you know, maybe maybe Calci can figure out a way to be Binance and we can figure out a way to be kind of hyper liquid in this scenario.
Speaker: Mm hmm.
Speaker: So you're kind of targeting like a different cohort of users in a way than they mostly have? Certainly initially. i mean, I think there's like a big cohort of a Polymarket International users that fit this fit this description.
Speaker: And ah here I mean, certainly there there are like elections people on Kalshi as well who who want to do international elections. But but yeah, I mean, I think right now these companies, their valuations are so big that they have been obligated to set as a target market like every young person in the United States or or maybe even in the world, which I just think is it's crazy. Like that's not an appropriate audience for for derivatives trading.
Speaker: Clearly, ah if if you're going to think about it that way. And, you know, Hyperliquid, what was their user count when they were um when when they first released their token, like probably somewhere between 10 and 50,000 daily users?
Speaker: um Like that that's a real it's a real Vanguard, but those guys are doing a ton of volume.
Speaker: Yeah.
Speaker: How would you, um like if if you only use ah Polymarket or Kalshi so far, how how would you explain um how Pascal is different?
Speaker: Oh, it's gonna feel like sliding into ah to a pool with a beer. on ah on a nice summer day is how i describe it. Or certainly that's our goal.
Speaker: No, I mean, ah it's it's DeFi, right? so So this would be the the first difference from Kalshi is that it's it's not custodial, it's it's on Solana. I think a big difference from Polymarket that you might feel is we've worked really hard to keep the kind of like matching engine, API, all that stuff very separate from the chain.
Speaker: like if you're if you're trading on Pascal, you never need to go look at Solana to figure out like what your position is or how much money you have. um And this sounds like a subtle point, but it's it's pretty awkward to be trading where you have two different sources of truth.
Speaker: I think you can look back, man, there's a great thread from this guy. I'm forgetting which Twitter account it was, maybe QuantXPT, someone who who had set up their trading operation on BitMEX and they were like accidentally pulling their balances from one of BitMEX's endpoints that like reads from their historical database and then they're trying to reconcile it against their live trading position. And they just like ate the spread over and over again and lost all their money.
Speaker: Like this stuff is pretty consequential. So if you if you have two different sources of truth that will cause your trading bot to do that, that's like a pretty terrible thing. um So I hope that people will find the API, you know, pretty responsive and and devoid of all those little kind of pitfalls. And then second, I think you'd notice that we have instruments that are not binary outcomes. So like if you want to go trade the anthropic pre IPO market, like let's say you want a short anthropic, you're like this thing. This thing is a bubble for sure.
Speaker: um You can go short perps, right? Like there are perps on Miter, there are perps on Hyperliquid. I think all the centralized exchanges probably have copied them now too. Like I'm fairly sure you can go to Binance or OKEx or Coinbase and do this.
Speaker: But um yeah you might get liquidated before the thing is even trading, right? Because they have to use some kind of like pre-market price as the Oracle.
Speaker: And, you know, they're giving people a lot of leverage. So if the if the price moves far enough, they they have to liquidate your position. i think we've seen this with a lot of like token pre-launch markets on Hyperliquid.
Speaker: I was, yeah I won't say which one, but i I was pretty short one of them and was liquidated just one of those wicks upward ah book before it opened. And so this is kind of like a necessity of perp design, makes sense. ah But what you have on Pascal is actually a fully collateralized linear instrument.
Speaker: So like usually a prediction market, it's like you buy one contract and the contract settles to $1 if the thing happens or zero if it doesn't happen. But for these instruments, which prediction market people like to call scalars, so we we're also calling them scalars, instead of either settling to zero or one, it's like there are a range of options. So let's say it's Anthropics valuation. We establish the range to be zero to four trillion dollars because that covers a large amount of the probability space. It's like if they IPO at two trillion dollars, then each contract settles to 50 cents. If they IPO all the way at four, then it settles to a dollar. they IPO at zero, then it settles to zero.
Speaker: um So you have this like sliding gradient and because it's fully collateralized, you can short it in as much size as you want. And there's no liquidation risk, right? Like you're, there's no way if you get liquidated because your counterparty is is fully collateralized. And so are you ah because we have this bounded range on the price.
Speaker: And that means if you want to make a high conviction bet about what's going to happen after IPO, this is this is a really attractive way to do it. Additionally, if you if you want to go long, there's no funding, which is another thing that's like kind of uncertain when you don't have a good mark price on a perp.
Speaker: um And in general, i think these instruments are really interesting for anything that doesn't have like a live underlying market price, but you want to make predictions about nonetheless. It's like you could imagine these instruments for like number of seats that a party wins in in the House or or the Senate or for GDP growth or, you know, any number of metrics that have kind of a continuous underlying index that you'll be able to observe eventually. You could build one of these contracts on.
Speaker: And to me, this is kind of like an obvious extension of prediction markets because the idea behind prediction markets is like, listen, there's a lot of things in the world. They're not all financial assets, but we can use the same techniques that we use to price financial assets to price these things.
Speaker: And that's good because it brings information. ah to the market and because some people really like this game of like figuring out how to value them person biased. But I think that's a pro as well. So um if you take that as the premise of prediction markets like this, just makes total sense. There are so many things out there that aren't binary.
Speaker: ah You can assign prices to them. People are used to trading things that have linear exposure like this, like perps. um So that's just kind of like the first the anthropic IPO, I think is is like the first stab we're taking at this.
Speaker: And you know if it's successful, if we can build a good UX around this, then what we'll probably roll this out to other other instruments as well.
Speaker: How does it work then if ah like theoretically something would go past the upper bound? Yep, so that that's like basically the trade-off. If Anthropic were to IPO at, let's say, $5 trillion and the upper bound is $4 trillion, the contract still settles to $1 trillion.
Speaker: Because... So it's like an option spread kind of where like the profit is limited basically. That's exactly right. It's an option spread. um I'm actually not sure where the term scalar came from.
Speaker: I think it might have been manifold markets, but definitely some prediction market people who are like, option spread? No. Scalar value? Yes.
Speaker: what made you What made you decide to build on Solana?
Speaker: Well...
Speaker: A little bit of self-hatred. No, we were naive about some things and it's been challenging, but I think it's actually probably still the best choice we could have made.
Speaker: so like Ethereum L2 bridging is the is the short answer. Like we wanted to settle every trade on chain. Like it's important for us that um there's actually a collateral check that's happening in a smart contract, especially after my experience on FTX. It's like this is like a genuinely good moral property of decentralized exchanges, like actually do the collateral check on chain, especially thing is totally collateralized. So this is this is trivial.
Speaker: So we knew we wanted something where we could settle a high volume of trades at ah at a decent rate. And so obviously Ethereum about one is out. ah The L2s have varying tradeoffs around like how easily you can do this. And then just the bridging experience for parking funds there is is was at the time quite challenging. I think now there are many more companies that have sprung up to address this. But my impression like mid 2025 was basically that this was going to be a terrible UX.
Speaker: And. and Yeah, I think I think that was that was really it. We weren't experienced really blockchain developers before we started doing this. Like all all my work had had been off chain. Certainly since then, we've learned a ton about obviously about how Solano works. And it's been a lot of fun, honestly. Like there are a lot more constraints than you would have on other changes on other chains. So it's harder to think through what the optimal way to design your smart contract is. But once you accept those constraints, ah you can do some pretty cool stuff that you couldn't do on other chains.
Speaker: And yeah honestly, it's it's been a roller coaster. It's been a lot of fun.
Speaker: How does um the settlement of events work? Because I think that is quite ah I actually don't know how Kalshi does it, but I think on Polymarket, it's done with this like UMA protocol thing or whatever, which has its issues sometimes. So how are you tackling that issue?
Speaker: Yeah, this is such an interesting part of prediction markets. Actually, I i had no clue... coming in how much of the P&L on Polymarket from various accounts was driven purely by games around settlement.
Speaker: Like there are traders on the platform where you can go and look at their profile and be like, wow, this is like a pretty high, sharp portfolio. Like whatever this guy's running, like it's working pretty well.
Speaker: Like what is the thing that it's that is working? You start digging in and it's like, They seem to be gaming the settlement process. Like that's crazy. There's like so much P&L that's being leaked through this.
Speaker: And I think from like a gamesmanship perspective, a lot of people actually appreciate this about Polymarket because they're like, oh, like when it's hard to trade on the fundamentals, I can actually just like play this kind of higher level of abstraction game where I figure out who's holding each side and like how much influence they have over the settlement process and I try and either influence it myself or I try and and tail them ah but needless to say a lot of people hate this because it's it's totally against the spirit of the instrument that you're trading it's like The European hell should come from being right about the about the event itself, not so much about ah buying influence with the the arbiter of the settlement. So it became very clear very quickly that there are like these two totally distinct axes.
Speaker: ah One is centralization versus decentralization, like this polymarked process for all its flaws, very decentralized. um And the other is trust versus untrusted. So I would paint them broadly as like Polymarket, very decentralized, very untrusted, Calci, very centralized, very trusted.
Speaker: ah Because their settlement process is like they write a contract and then they try to abide by the contract and you know some precedents are set, etc. etc So what did we decide to do? we we looked at this like we're definitely not doing some kind of decentralized scheme like this. is This is a terrible user experience. And if you want these things to actually trade um like financial instruments, like you you basically can't do this.
Speaker: So for now, we are we are listing and settling all the markets ourselves. Like we try to write pretty reasonable, tractable markets because of this.
Speaker: So one example is we have a market on PolyMarket's uptime. Like they've had all these but all these issues where, you know, their their API ah goes down, their website goes down. And so we're like, all right, this this would be an interesting thing to have a market on.
Speaker: ah But how do we settle it? like there are all these different definitions of uptime. So we literally in the rules have a script that you can run and the script will go and like look at their status page. and It'll be like, we're counting these incidents. We're not counting these incidents based on a clear...
Speaker: criteria so you can run this algorithm at any time to understand like how would a status page that looks like this influence the settlement of this instrument um so then obviously in those cases like we just run the script when it's time to settle the market because that the end date has been reached and then we push the settlement on chain it sells all the positions and people are people are happy ah In the future, i think you can imagine there's kind of a trade off right between how legible, how legible your market and settlement process is versus how interesting the events that you can cover are.
Speaker: um And there's there's definitely an art balancing those things. I think i don't want to say too much, but we have some plans for improving this in the future and kind of widening the aperture of the markets that we can list.
Speaker: And I think from our perspective, it should be an engineering challenge. I think it's only you can only really scale so far if you make this like hire 50 college grads as like your market listing interns and like try really hard for them not to make mistakes. So just like tell them every day, please don't make a mistake. Like that there's only so far you can get with that approach.
Speaker: So. um I think it's kind of a meme, so I'm i'm reluctant to say it, but I do think LLMs can play a huge role here. And, you know, I think the bad way to do it is is actually quite bad, but they have some desirable properties like ah even without verified inference, which we're now seeing some research come out of Anthropic for doing like CK verified inference, which is super cool. You can have like plausible inference from an open weight model that's that's actually Like it's pretty easy to look at the results from an open weight model and say like, are these results plausibly the output of the model, given the inputs to it? Because I know the weights. I know kind of like how it's sampling, like how out of sample is this answer?
Speaker: So if we are over here, like putting some rules and and criteria through a model, taking the answer and then posting that on chain, people can look at that and say, like, is this reasonable? And then second is that the LLMs are obviously ah completely impartial in this process. They might be biased or wrong. So that's kind of the engineering challenge to get around. But they are not partial. like they They don't hold shares in any either side of the outcome.
Speaker: And I think that's actually probably more important than being perfectly accurate. ah But luckily, they're also quite accurate, like with with the tool calls that they have now, um maybe even with with a bit of a custom harness, especially for events where you can feed in a permissible set of sources. It is not too much of an engineering challenge to say like, hey, for credible reporting, you can only check these news outlets.
Speaker: And you're going to you're going to check these news outlets and you're going to determine whether US has reached a trade agreement but with this other country. And a benefit of this also is that anyone who is reading the rules and understands this process for how they're eventually going to be settled through this through this LLM can actually just go run it themselves with different hypothetical inputs to understand how it would settle.
Speaker: um So I won't say too much. These are kind of some some raw thoughts, but I think there's a lot that can be done there to actually automate this whole exhausting process that happens right now around market resolution without introducing too many additional risks.
Speaker: It sounds very interesting. um How do you go about bootstrapping liquidity? And also, can you talk a bit about your your fees and how they work? Yeah, fees on prediction markets are really funny. they're um So that this is standard on every prediction market, and including us. The fee is expressed instead of like price times size times some percent.
Speaker: The fee is price times one one minus price times size times some per percent. And if you come from like a perpetrating background, you're like, what the hell? Like, why would you do this?
Speaker: And the answer is it's it's kind of required because um People get this wrong a lot when they're trying to compute volume as well. It's like if you have a binary outcome and you're quoting volume as price time size, then you get into a position where if you if you state something in the affirmative,
Speaker: like, will the Democrats win the Senate and it's trading at 99 percent? ah You get a totally different volume number than if you had taken the same question, done the same trades and just stated in the negative, like, will the Democrats not win the Senate?
Speaker: Then suddenly it's trading at one percent. So like clearly neither of these is is a good measure of of what's actually happening. So for the fees, ah taking price times one minus price is a good way to make it symmetrical around 50 cents.
Speaker: And it also um That value is kind of precisely the theoretical variance of the outcome if the price is correct.
Speaker: So it ends up being kind of a way to to charge fees in proportion to how much risk is actually being exchanged. um So interesting quirk of prediction market fees.
Speaker: Our fees are lower. Obviously, i think Polymarket just increased their fees. Calci has pretty chunky fees. um One effect that this has on people's behavior, which you can see when you're trying to bootstrap liquidity, is actually like most these markets, obviously they trade zero to one and most of them have cent ticks, which is actually pretty high, right? Like if you're if you're trading like a 50 cent outcome and there's like a one cent tick, like already that's that's two percent just gone in the bid-ask spread.
Speaker: Now, increasingly, we're seeing point one cent ticks, which I think is a little bit better. But, you know, there's there's a big cost to crossing the book. And then if you add on a pretty high fee on top of that, like Cal cheese is 7%, Polymarket US just rolled out a new 9% starting tier, i think. And then Polymarket International has anywhere between like 3% and 7%, depending on the market type.
Speaker: We are at 1% or 2%. We have some plans to maybe cut those even further now. um Yeah, so if if if you're looking at at fees of that order of magnitude and ticks that wide, it's like really hard to do fine, fine-grained price discovery because even just crossing the spread is is costing you a ton.
Speaker: So how does this affect people's trading? I think if you look at pro sumer kind of traders and perps, which you would know very well, you guys probably have a lot of data on this, like even the the like real upper end of the of the trading spectrum, people are mostly takers.
Speaker: Like it's not it's not very common for people to be placing maker orders in the order book. Whereas in prediction markets, this is totally different. Like many of the power users are like 80% maker orders.
Speaker: And this results from these transaction costs that I'm talking about and also from Polymarket's equity rewards program where they they reward you for being top of book.
Speaker: And So this this apologies for the long winded answer. This is all to say that this this stuff is intertwined. So like fees, bootstrapping liquidity, ah how big the spread is, the tick size, like all of this stuff contributes to like, how do you think about starting a market with liquidity?
Speaker: And yeah, maybe I'll contrast our approach to to Polymarket's liquidity rewards approach, because I think they copy their liquidity rewards from DYDX. Like if if you go read the first version of their docs, they're like,
Speaker: these liquidity rewards are similar to the formula on DYDX. Like click here to read the DYDX talks about them or whatever. And um so we had some experience with these for our time at DYDX. And I think they're bad for for a couple of reasons.
Speaker: First, especially in PolyMarkets implementation, because the ah the tick size is so big, right? It's like these one cent ticks on a lot of markets are 0.1 cents, whereas in Bitcoin, you're like a much smaller fraction of the total price. Because the tick sizes are so big, ah it's actually extremely easy to just hide behind the person who's at the front of the order book.
Speaker: Like you come to the order book, you see someone who has posted the bid-ask spread, and then you're like, great, I'm going to put my order behind that person. And then if someone trades against them, like probably they're not going to hit me and i'm going to be able to cancel and like back off. So I'm going to farm these liquidity rewards and also I'm not going to take any risk of trading like this is great for me.
Speaker: um And I think. You know, like you can also you can also run into problems with that, like if you do get filled. like conditional are getting filled, it's much worse to be at the back because it means there's someone who's like really eager to sweep out a few price levels. so that could be bad for you.
Speaker: But like in general, there's so many order books. There's such sparse attention to them. They're paying out liquidity rewards left and right on these books. Like it's not that hard to kind of just camp out and farm these rewards without bringing a lot of value you to the market.
Speaker: um So we knew we didn't want to do that. And also. i think it kind of incentivizes the exchange to say like, well, we did our part. Like weve we put these liquidity rewards out there and like now we can just like not worry about market structure, which is obviously not the approach that that is interesting to us. So we have a trading desk internally and our trading desk obviously quotes the majority of these markets.
Speaker: I'll say we don't really see it as an enduring business line. Like our our hope is kind of like similar to HLP. Initially will be a large percent of the market share. We'll be able to bootstrap markets that are interesting enough for other makers to want to come in. We're going to make it really, really easy for them ah to do that. And then our desk will like move off to the longer tail markets because there are just so many things to quote. You kind of like bring in specialists market by market. You move out to longer tail things and that's how you kind of bootstrap activity on the exchange.
Speaker: Increasingly, ah So we pay we pay a make rebate of every fee instead of having these liquidity rewards and we quote the markets ourselves. So if you come and and you want to place orders, you only get paid if your order fills. But if your order fills, you get a percent of the taker fee. And that kind of cuts against this dynamic of like trying to farm liquid liquidity rewards by not getting traded against.
Speaker: um So that's that's first step. Then the second step to bootstrapping a market is like you need people to come and trade. So we just ran this this midterms ah trading rewards program where, you know, we we basically made it extremely cheap for people to trade. Like you come trade midterms on Pascal, you'd have decent liquidity and the fees would be low or after rewards, maybe negative.
Speaker: um We're turning those off now and we're starting to think about what to do to incentivize liquidity going forward from third parties. And. You know, a lot of people have asked us about liquidity rewards.
Speaker: i I really don't like the poly market style of doing it or the original DYDX style. um I think. i I don't want to say too much just yet, but I think we can do something much better. And then given how people have been conditioned to place on these markets, I think it might actually be more interesting to reward liquid liquidity ah by taking.
Speaker: Like if you if you imagined, ah you know, taker orders are going to are going to come through the book and you know, they're going to be noise. This incentivizes liquidity because you want to go compete to to be at the front of the book to get these good fills.
Speaker: But then you also want to have an idea of like what the price is that you're getting out of the way bad fills. And i think a model like this, I'm not sure exactly how we would want to implement it, but I think a model like this could be a lot better because it it would really incentivize the makers who are also interested in knowing what the price is and create a competitive environment where where you're trying to get filled as opposed to trying not to get filled.
Speaker: I think that was a good, interesting answer. It definitely makes sense. um I guess it's kind of slowly time to wrap it up because the episode is already a bit long, but I really enjoyed this one. I think this was very interesting. I really enjoyed listening to to your background and what you've done so far in crypto and also about what you're building now.
Speaker: And um it's it's a bit out outside of my... usual ah knowledge scope of it, I guess, since it's more about the the infrastructure, market making side or whatever. But it was very interesting and I hope the the listeners agree with me here. So I want to thank you very much for for your time and for coming on.
Speaker: Thank you so much for having me here. It's been a pleasure.
Speaker: And thank you everyone for listening.






