To shed light on how various protocols address this dynamic aspect of the market, we have initiated a series of collaborative discussions dubbed Volatility Talks. The first of these was an insightful conversation with CVI where both sides share their perspectives on navigating volatility within the DeFi ecosystem.
Join us as we delve into the projects and strategies for maximising trading opportunities amidst market fluctuations.
[TRANSCRIPT START]
Baz — Bumper
But today is the first of what we’re calling our volatility talks. What I wanted to do is organize a bunch of collaborative AMA’s with different protocols across the Defi ecosystem. And in this case today we’ve got two protocols who either facilitate trading opportunities or approach volatility, specifically in the protocol. I thought it’d be good to host this discussion and unfold some of the detail there as to how volatility plays a role in the protocols and how traders can maximize opportunities from utilizing it, trading it, and building positions from there. So I think maybe if we start with two introductions, we’ve got speakers on the line. Moshe from CVR finance and Sam from bumper.
Baz — Bumper
So if I bring you up to the mic and if you could give us a bit of an introduction on you and your background, and then we’ll take it from there.
Moshe — CVI
Sure. So I’ll go first. So great to be here all. My name is Moshe, I’m the CEO of CVI and one of the partners, entrepreneur for the last 15 years, been co founding companies in various areas, including tech, crypto, Internet, etcetera, and been leading as the CEO of the company since last year. And. Yeah, excited to be here with you guys.
Baz — Bumper
Cool. Thanks, man. Thanks for joining. We’ll dig into a little bit more about CVI in a second. Sam, if you could give us your intro and background.
Sam — Bumper
Yeah, sure. So good after one. Sam Brooks. I am originally an electrical engineer and spend a bit of time as a professional engineer in Australia and also working in a couple of Australia’s two largest banks, commercial banks, retail banks, got into blockchain, must have been in 2014 or 15. Once I read about Ethereum, and once I read about Ethereum, went back, read the bitcoin white paper, and then suddenly realized just how interesting this technology was. Anyway, fast forward a few years and I came up with the design behind the bumper product, teamed up with the CEO and the COO, and now I’m CTO, leading the development and research and engineering efforts.
Baz — Bumper
Good stuff. Thanks for the intro. And then right on time, I see Jonathan was just able to log in. Hopefully we’ve now got you. And Mike, could bumper host add Jonathan a speaker? Let’s see if we can get you onto the audio. The other way is, Jonathan, could you do the request to speak? There you are. I think you’re done.
Jonathan — Bumper
Yeah. Hopefully you can hear all hear me. Yeah. Were we doing kind of loud and clear? Were we doing introductions? I’ll take.
Baz — Bumper
Yeah, just personal introduction. So, yeah, you can introduce yourself, Jonathan. That’d be great.
Jonathan — Bumper
Okay, well, just to give a very short potted history, I studied artificial intelligence and psychology in 1995 at the University of Nottingham, which at the time had the largest neural network, and I studied under Sir Nigel Shabbolt. That went on to create the Internet of things. Subsequently to that, I’ve started, I think it’s in the region of 28, 27 different businesses. Most of those didn’t really go anywhere. Some total failures and a few notable successes, such as Switch, which is the UK’s 60th fastest growing company, Launchpad, which was one of the first property crowdfunding platforms.
Jonathan — Bumper
And then obviously kind of moving more from fintech into crypto starting index, and then obviously most recently bumper, of which we raised a $20 million kind of research and Development fund back in 2021 in order to produce what we think is a novel and suitable candidate for on chain kind of price protection.
Baz — Bumper
Exciting stuff. Good to hear. Thank you. So I think what we’ll do is dig into a little bit. It’s kind of a good segue into the two protocols themselves. I wanted to start with introducing the people and then the protocol, and then we’ll get into some questions about how volatility plays a role, maybe. Marcia, if you could give us a bit of an intro into what CVI finance is, and then we’ll go from there.
Moshe — CVI
Yeah, sure. So CVI is the crypto volatility index. We’ve been around for almost four years already. We started with the idea of bringing the implied volatility trading capabilities that Radfi are so much enjoying from to the crypto space back then four years ago. And over the years we have been building various versions of the same product, improved versions of the same product, which is a 30 days implied volatility index on bitcoin and ethereum. And through our platform, you can actually trade against the volatility of bitcoin and ethereum together. The counterparty of the trades is what we call the tetavolt. The tetavolt is the liquidity that used for the trades, the counterparty, so on. The tetavolt depositors can deposit their funds and enjoy from funding fees that traders are paying when they open their trades and holding their trades open.
Moshe — CVI
And this is the ecosystem that we have been building for the past four years. And yeah, this is more or less about CVI as a background. We have some interesting developments coming up. I’ll talk briefly about some of them during these x spaces. Yes, that’s CVI.
Baz — Bumper
Cool. Thanks. Thanks. Just as a quick couple of follow on questions for the audience, how long has the protocol been live kind of trading volume are you seeing?
Moshe — CVI
So the protocol has been alive for four years already? More or less. We have released a very significant version on December, which now allows traders to trade with leverage of up to almost 50 x and some other features that brought up the volume to be much more significant than we had in the past. Since then, we’ve seen cumulative volume of around $20 million. And on a daily basis it changes because obviously the market is in some cases very volatile and creates trading opportunities, in other cases less. But we’re seeing anywhere between one hundred k to two million dollars volume daily. Yeah, awesome.
Baz — Bumper
That’s good to hear. Thank you. Good for the intro. And I think jumping over to bamboo, I don’t know if, Jonathan, you wanted to intro and then throw to Sam if there’s anything that you wanted in to add or build on.
Jonathan — Bumper
Yeah, I’d love to jump in there. Moshe, can I just ask you a quick question there? So you talked about a kind of leverage element. Could you just dig into that a bit more for me? There is that kind of leverage options you’re talking about there.
Moshe — CVI
No, it’s not a leveraged option. The way CVI is built is that we have the index, the CVI index, and as I said, we have traders from one side and the tetavall from the other side. So you are not. I mean, the calculations are obviously based on the option prices. And we are using the black Scholes formula to calculate and bring everything on chain in real time. But then the traders and obviously the depositors are agnostic to the option prices. All they see is the actual outcome of our calculations, which is the CVI index. And when we’re talking about the leverage, it’s very similar to other protocols like GMX. Right. It also works in the same way where you have liquidity and traders from the other side so people can open their position with leverage. And, yeah, this is how it works.
Jonathan — Bumper
They’re kind of a GOP pool on the other side that basically makes that leverage. Yeah, got it.
Moshe — CVI
Yeah, yeah, exactly.
Jonathan — Bumper
Cool. Okay. That’s really cool model. And congratulations. That’s really good traction.
Moshe — CVI
Thank you.
Jonathan — Bumper
Just to kind of speak about bumper for a moment, then. So we took the decision three years ago that we thought there might be a kind of more efficient mechanism to serve up protection on chain, obviously, kind off chain. I think black Scholes and option desks have done a very good job both within crypto and traditional finance. But there does seem to be some kind of problems on chain. You know, you kind of look at the prices of lyra and hedgehog and the other kind of derivatives of kind of defi on chain options. They are more expensive. And to us, the fundamental issue was that certainly, even with deribit, which has the deepest market, is it’s very difficult to get the option size and term that you really want. So they’ll often be very lumpy.
Jonathan — Bumper
It might be, well, there’s an option for two days that you can buy and then an option, you know, for maybe five days after that you can buy. So, to put bumper very simply, then imagine a kind of automated market maker that can serve up any type of option that you desire. We do not use black Scholes as the underlying formula, but we spend a huge amount of energy and an in depth kind of agent based model which included various kind of AI streams in order to build a protocol which actively rebalances between a volatile asset pool and a stablecoin asset pool. And on the one side, obviously, the people protecting their asset, be it bitcoin or ETH, will pay a premium that is then paid out to the stablecoin providers on the other side.
Jonathan — Bumper
But the real secret sauce is a rebalancing mechanism which uses various ratios that are viewing the market in order to trade out some of its volatile assets and buy more stable coin assets as the market moves. At the moment, in our fairly embryonic history of, I think we’re coming up for six months now, and 2 million traded notional value is on average, about 30% cheaper price point than derevit. So what we’re finding is that bumper, we hope, is an extremely efficient protocol in terms of serving up the option dimension that you’re looking for, but also bringing a massive efficiency on price to the market. And that’s something we hope to kind of bring to the wider market as we ramp up now.
Baz — Bumper
Sounds good. And by means of kind of introduction on the protocol, is there anything you wanted to add, Sam, or should we move into questions?
Sam — Bumper
No, let’s go straight into questions.
Baz — Bumper
Cool. Yeah, sure. So, I mean, volatility talks was the name of the topic today. I think one of the things I wanted to dig into was, first of all, from CVI side, if you could tell us a bit more about the CVI index itself as a metric, and how volatility plays a role in the protocol, that would be super helpful.
Moshe — CVI
Yeah, sure. So this is a very good question that I always love to answer, because it helps people understand how to trade CVI, and not only how to trade, but how to look on the market from the CVI perspective. So, as I started to explain before, we calculate the CVI value using black Scholes and option prices, and then the outcome is a number, and the number represents the market expectations. So if I take right now as an example, so the CVI value is somewhere around, let’s see, 57, let’s say, 60. So that means the number 60 means that the market expects that the price of bitcoin in Ethereum will move by 60% in the next 30 days, annualized. Okay.
Moshe — CVI
This is the bottom line of what this means, and if we want to understand what is the market expectations for the actual next 30 days, we need to divide 60 by square root of twelve and which is, I think maybe 15, I don’t know, around that. So these are the market expectations. This is what the market actually expect that the movement in the market will be. And the market moshi, that could be.
Jonathan — Bumper
Up or down, right?
Moshe — CVI
Yeah, yeah. It can be up or down, obviously. So those are the market expectations, right. And they change according to different events and according to different things that happening. But these are the market expectations. So usually when the market is very steady and low, you will see the CVI value go down. And when there are either big drops or big spikes, you will see the CVI value going up. So this is more or less about CVI and what it represents.
Baz — Bumper
Is that metric completely proprietary to you guys, or is that something that you pull in or have kind of calculated based on other metrics?
Moshe — CVI
No, no, we calculate it ourselves. Obviously, we got the inspiration of how to calculate it from similar tratfi products such as the VIX. And actually the originator of the VIX, Professor Dan Galai, is one of the advisors of the company, and helped us through the initiation of everything here. So we haven’t invented the wheel on that front. We have invented the wheel in crypto. This was very hard to build everything for bitcoin and ethereum. It was a lot of adjustments that we needed to do, but we calculate everything from maybe 90% of the options exist in crypto on bitcoin and ethereum, and we do everything on chain in real time. Yeah, yeah.
Baz — Bumper
Very cool. That’s good to hear, I think. Very good too. Yeah, it’s a really cool metric.
Jonathan — Bumper
And could I just ask a quick question, Moshe, then? So I’m fascinated by what you’ve been talking about. How is the prediction of, say, that 15% volatility actually, how does it bear up in backtesting or just, you know, just general kind of correlation with the actual volatility?
Moshe — CVI
So, you know, it’s very hard to backtest this, because every day the number can be different. So it really depends on the day. Like in 30 days, you’ll probably have 30 different numbers and that you can back test for the prices of Ethernet Ethereum. The market anticipations are pretty accurate. One of the things that we have seen is that, for instance, back in January or even December, before the ETF approval of bitcoin, one of the biggest things that happened is that, I think a month before the ETF approval of bitcoin, the CVI value was very low. And some people realized that this cannot be the case, like, as we progress, as we get into the due date. And they realized that, and they actually opened their positions before.
Moshe — CVI
So we saw a massive increase in OI because people realized that CVI value will only go up until we get to the ETF approval. So obviously, those times where you have very big event, very major events there, the market can be a little bit crazy, and the value won’t be necessarily representing the actual movement in the next 30 days. But in times like this, I think that the market can be pretty accurate in terms of the movement.
Jonathan — Bumper
Yeah, got it.
Baz — Bumper
Cool. Yeah, sounds good. And jumping across to Sam, I wondered if you could give us a bit of insight as to how volatility plays a role in bumper itself.
Sam — Bumper
So it plays a very important role, as it does in most other constructs for pricing risk, just to set the scene effectively. What bumper does is it computes the premium for all positions in the protocol incrementally. So as volatility is measured over time, the premium that’s computed within the protocol will go up and down and it accumulates indefinitely, while there’s at least one, what we call a taker position in the protocol, and is then transformed into an individual’s premium based on things like their. Their term, in fact, the exact timeframe that they were in the protocol, their chosen floor price, which is equivalent to a strike price in black Scholes, how much, of course, they deposited in terms of their asset, etcetera.
Sam — Bumper
So there’s a huge sort of conceptual overlap between the way bumper works and the way black Scholes works, which is, I think, quite interesting when you think about it, considering we only really became, admittedly, only really became aware of the black Scholes equation about halfway through the design process. And at that time, it became clear that our design had kind of resolved the same fundamental inputs that black Scholes uses in terms of volatility itself. So the way bumper uses volatility is that it, as I said, measures price activity over time and effectively operates more frequently. So the premium will compute itself more frequently the more volatile the measured price of the underlying actually is. And so that price risk factor, as we call it in the design, it goes into the computation of the premium as one component.
Sam — Bumper
That price risk factor also goes across to develop what’s called a liquidity risk factor, which is the second component of the premium. So there’s a couple of things going on there. So whilst there is a lot of overlap with black Scholes in terms of the way, sort of at a high level in terms of the inputs that go into both methods. One of the, one of the, or, I suppose the first criticism of the black Scholes equation is okay, parameters, which establish how to compute this fair price of a european put option, and.
Baz — Bumper
All.
Sam — Bumper
Of the choice associated with whether or not a buyer and a seller of that put option accept all. The choice related to whether they accept that price is sort of distilled down into this, into the volatility. And so that naturally drives the question, well, how do you agree on what the volatility should be? Because everything else is naturally agreed. They’re effectively the parameters of the product or of the contract between the buyer and seller. But how do you agree the volatility in one of the first obvious criticisms of the basic black Scholes equation, is that okay, well, it uses static volatility, whereas as we know, volatility is dynamic, over time, it changes. And of course, there’s been decades of research about how to do it differently.
Sam — Bumper
So in terms of bumper and the way that works, we do it differently by accumulating the premium over time for the measured volatility. And that goes into, of course, as I mentioned, derive a price risk component to the premium and a liquidity risk component to the premium. And that also, in fact, goes into the rebalancing mechanism that Jonathan, based on the protocols understanding of what the price is doing, is it high risk or low risk? Does it think it’s going up or down? The rebalancing mechanism in between the two pools will dynamically hedge in between both pools of value, effectively to maintain. Maintain the current value, so that.
Baz — Bumper
The.
Sam — Bumper
Liabilities which exist for all positions can slowly decay as the premiums are accumulated and applied to those liabilities in real time. So that’s how, that’s a brief description of how volatility is treated in bumper.
Baz — Bumper
That’s definitely insightful for me, and I work for the team, so I’m learning something every day which is good as far as black shells go, it’s come up quite a few times. Moshe, on the CVI side, am I right in thinking that you put black shells in as a metric into the CVI index? Is that right?
Moshe — CVI
Yeah. I mean, I’m not a data scientist, but right from above, we are using a form of reverse engineering of the black Scholes model. Then again, very much like the VIX, and the model is using the prices of the options, which are calculated by models like the black Scholes. And this is in order to calculate the expected future volatility in the market. So, as I said, we’re inspired by similar implied volatility trading products on the tradfi, and this is what we’re using to calculate the CVI value.
Baz — Bumper
Yeah, gotcha. Makes sense. Whereas on the bumper side, it’s much more of like, I guess, more like a competitive reference point. It’s not used in the protocol, but it’s something that we, I guess, compete and try to become more efficient than an existing tradfi metric.
Sam — Bumper
That’s true.
Baz — Bumper
Interesting. Different view.
Sam — Bumper
That’s. And that’s just to add to that bazaar, you know, when we, I mean, first of all, I recall started last year when we first turned on the simulation. The, it was quite astonishing to see the correlation between the bumper price as it was resolving over time, accumulating over time, and black shoals. It was, it was very, very interesting to see that sort of numerically manifest itself in the results, because conceptually, we’d always had this understanding that, you know, there was a lot of, phenomenologically, I suppose there was a lot of overlap between how the two pricing methods work, but kind of, if you go one layer down, they do work quite differently in terms of that accumulation and measuring the price risk over time.
Sam — Bumper
But subsequent to that, the simulation was incredibly useful in tuning the protocol parameters to actually, over a range of volatility scenarios, over a range of time periods, and different TVL levels and concentrations of risk between the pools, etcetera. To actually tune the protocol, to perform in between what a pure black Scholes equation would have developed as a price and what the realized market prices were for, put options. That’s exactly how we did it. We wanted to be kind of right in between, halfway in between, in fact, black scholes as a baseline for a completely fair pricing for risk, and what we see on Darabit. And that was the other, that was the sort of market benchmark that we took.
Sam — Bumper
And we got a data feed from derebit and replayed all of the pricing action over a period of four years and tuned the protocol to perform exactly within those two numbers.
Baz — Bumper
Gotcha. Yeah, makes sense. So then, switching gears from the conversation and putting both of these protocols in the hands of a trader, what are the best ways that a trader can utilize the protocol to, I guess, like ride the waves of volatility, trade volatility, or kind of, I guess, harness and beat volatility? Maybe if I jump back to moshe first?
Moshe — CVI
Yeah, sure. So we’re also very open about it because we want to help educate the market. I think that implied volatility is not yet utilized as it should be, very much not compared to tradfi and similar volatility trading products. So we invest a lot of resources in helping educate the market and help both traders and liquidity providers to participate in this amazing ecosystem. Because I really think that, you know, in opposed to a lot of different assets, volatility trading, you can do like every day, the entire year, whether it’s a bear market or a bull market. There is also, there is always a room to trade volatility if you understand how this magic works.
Moshe — CVI
And from the eyes of trader, I think that the main trading opportunities that I see is one is a very straightforward range trading in opposed to a lot of different assets. The implied volatility. The CVI index, the implied volatility in general, it lives within ranges. Every day, week, month, there are certain ranges that this CVI value lives. And then when it goes down, it bounds to go up, and when it goes up, it bounds to go down at some point. So I encourage traders to learn how to range trade CVI, understand the ranges and then take advantage of, at this point, the lower end opportunities before it goes up. I think that we have seen a lot of trading activity when CVI value went up, and now when CVI value is lower, we see less activity.
Moshe — CVI
And it doesn’t make sense to me because I think that right now is the opportunity to trade CVI when it’s lower, when it’s higher, I mean, obviously it is more volatile and there are certain opportunities, but when it is low, at some point it will go up. So try to identify those ranges and take advantage of it. So yeah, one is rain trading, the second are events. So we have identified certain events that some of them were published on our Twitter and Telegram communities. One of them, for example, is the FOMC. So we’ve seen that in four out of five previous FOMC meetings, the CVI value went up the days before or during the FOMC. Also important to note and to understand that usually this is the case.
Moshe — CVI
Like before and during certain events, the volatility will go up and then at the end or toward the end of this specific event, it will go down. So identify those events. I mean, you can investigate research, do your research, go back and see which events are causing the volatility to spike. In our industry, FOMC is one of them. And then plan your trades ahead. You can use limit or market orders on our platform, plan your trades ahead and take advantage of specific increased volatility events such as the FOMC. And the third one is a more complex event. But CVI, as I already explained it, represents the market expectations. So usually when the price movements are bigger than the market expectations, for instance, in cases where there is a massive price drop. Right.
Moshe — CVI
And then again, I’m taking you back to the beginning for a conversation of what the value of CVI actually means. So a movement of, an expected movement of the CVI value of 60 represents an expected movement of, I don’t know, like in the next 30 days of around 15 or 17%. That’s a lot. So it needs a very big movement to get even. If bitcoin drops by two 3%.
Moshe — CVI
CVI value can still remain the same. But in certain events where bitcoin or Ethereum price will drop massively, then CVI can also be used as a hedge. So if you are expecting a specific event or specific times where you need certain hedging, then CVI can definitely be helpful. Then again, events such as the upcoming ETF approval or ETF rejection of Ethereum, the 25 may is an amazing opportunity to either hedge yourself or profit from volatility, whatever way you’re looking on it. But those are the three main things that I see. One, range trading, two events, and three hedging.
Baz — Bumper
Yeah, the events. One sounds like a really easy and obvious play, doesn’t it? Long in the lead up to it and short and the follow up from it. So there’s some serious not financial advice alpha being shared there. Thank you.
Moshe — CVI
Yeah, yeah. Also, yeah. Saying not a financial advice, not financial advice.
Jonathan — Bumper
Never advice.
Moshe — CVI
Never.
Baz — Bumper
It never is. It sounds good. And then Jonathan on our side on bumper, if you could give us a bit of a flavor as to some of the opportunities that a trader has to utilize bumper in their trading strategies.
Jonathan — Bumper
Yeah, sure. I’ll try to keep it as succinct as I can. I think the main mission statement we had when we started out was, can we find a way that democratizes everyone to be able to take protection? Right. And that’s what bumper has successfully deployed. That just in, like, I think it’s like four clicks, you can protect your ethereum. A percentage price it is now. So, for instance, if you protected 95% of the price of Ethereum today, and we’re talking Rapstate ETH, which has a higher price point than traditional ETH, then it’s, we’re talking about an asset which is worth approximately $3,500. The price for one week is $35. So that’s 1%. So if you’re in that predicament where you think the price is going to drop, bumper is an incredibly efficient mechanism to do that, and it’s available to anyone.
Jonathan — Bumper
It’s ridiculously efficient to use in terms of the UI. So obviously that opens it up to traders to kind of utilize that, basically that arbitrage of volatility and profit from that. I think the key thing that bumper has always really been interested in is, let’s say you have a bitcoin. You send that bitcoin to the bumper protocol, and you protect that bitcoin at 95%. What the protocol returns to you is bumpered bitcoin. Now, this is the goal that we think is the real game changer. So what you now have is a version of bitcoin where the downside volatility has been removed.
Jonathan — Bumper
You can use that to open up, say, a collateralized deposition, leverage longer leverage, short some kind of risk on derivative, with a certainty that even though the price of bitcoin might drop, say, 20%, your bumper bitcoin is only ever going to drop 5%. And you’re paying a charge for that, right? Paying the premiums. And as far as our modeling calculations forecast, you’re going to make a handsome profit by taking that protection because of the efficiency of the pricing of the bumper premiums. But what it means is that we’re ushering in a kind of era where liquidations could be managed by a very efficient mechanism.
Jonathan — Bumper
And we think that is an extremely powerful tool, not just for other protocols, but obviously for traders and treasury funds and anyone that wants to take a hedge position, but then re hypothecate and utilize the derivative token in other protocols. And, you know, that’s something we’re really interested in getting ourselves integrated into.
Baz — Bumper
Gotcha. Yeah, big one. And I would point anyone that’s listening to an article on our blog where we wrote six new trading strategies where you can utilize, I guess, different strategies to beat the market. And there’s several different ways there, from earning yield to hedging the drops, or riding the waves up indeed, in bumper as well. So there’s an article on the bumper blog there, just to close, I guess. Looking forward at roadmaps and direction of these two protocols, is there any exciting nuggets of alpha or insight you could give us on what’s coming next down the pipeline? Starting first on the CVI side, and then I’ll come to bumper.
Moshe — CVI
Yeah, so I mentioned before that we launched the fourth version in December. And it was a pretty big release that we have been working on for a few months. And I think that right now the CVI product is pretty much in place where were expecting it to be. The funding fees are in a place that we feel comfortable with and the leverage and the UCBI index and other things that we released. Having said that, we do have a very big change that we are going to release in a few weeks time. It’s the short functionality that was available only for allow listed market makers since the launch of the Ford version, and it will be exposed to all of the users in a matter of, I guess maybe two weeks, three weeks maximum. This is what we expect.
Moshe — CVI
So this is one thing that we’re going to release and it’s a big change, and it will allow traders not only to long volatility, but also to short volatility with leverage. And I think that then after, we’re not expecting major changes to the product, but we are working on some collaborations, interesting collaborations, and some future plans to expand our offering, which I cannot expose right now, unfortunately, because it is still not official, but we’re working on something big. But yeah, short term is the short. Yeah, and the longer term will be announced very soon.
Baz — Bumper
And on the bumper side, maybe to you, Jonathan.
Jonathan — Bumper
Yeah, I think for us the roadmap is probably a massive step change up. The smart contract work was obviously a kind of challenging piece of work, but what we’ve been working with in parallel are three, at least three AI stacks that we think are absolutely game changers. And, you know, all kind of. I want to explain what those stacks are so that people kind of have a better picture. So let’s start demystify this kind of whole AI thing. So the first stack is a 70 billion parameter large language model that we use to feed hard financial data. So this is numerical data. Now, traditionally, llms have not been very adaptive to numerical data, but we’ve been able to have success with training, proprietary training of data and fine tuning that with something called reinforcement learning from human feedback.
Jonathan — Bumper
So if you imagine you kind of do a backtest, you feed tick data of, say, the price of bitcoin over the last kind of decade, that be the optimum. And with that data in its kind of neural network, you then ask it to make predictions of 1 hour ahead, and when it gets it right, you feed back that, you reward the LLM. And obviously if it kind of makes an incorrect prediction, you kind of feed that back. And so we’re starting to have extremely powerful responses come back with that. And there’s a very interesting approach we’ve taken with what we call kind of in context submissions of data. So this is where in the text prompt to the AI, you can actually paste in data. And those context windows vary between 128,000 tokens to a million.
Jonathan — Bumper
With Google Gemini Pro, even that is too small if you want to kind of feed in tick data. So we’re using something called Rag, which is retrieval augmented generation, which is an extremely efficient way to transform traditional databases into vectors. And that format seems to be extremely powerful when you feed it into llms. And this is only technology that’s been around in the last two or three months. And we’re really proud to be embracing that as a separate stack. We have an 8 billion parameterized LLM, which is ingesting vast amounts of data from Twitter, Discord, telegram channels everywhere, where you would imagine this kind of commentary on kind of crypto. And the purpose of that is to build up kind of general single signals and whether the market is positive, negative or neutral.
Jonathan — Bumper
And then the final stream that is actually looking, it’s going to have great price. And we think we have kind of proprietary oversight on this at the moment, is taking existing bitcoin pricing. So historical data, scripting up graphs from that, automatically scripting up labeling of that graph. So there we’re talking about resistance support levels, MACD, RSI, Bollinger bands, volume levels and so forth, and feeding those raw images into a lava model, which is a kind of cross between a visual model and an LLM. And we’re starting to find some very interesting kind of results coming out of that. And so what we have is an oracle, which sits in the middle of all of those, where the signals feed into it.
Jonathan — Bumper
And it kind of tweaks which ones it listens to the most under certain market conditions, as it’s learned through its reinforced kind of feedback. And what we are predicting, this is something that we’re working towards, this is our mission statement, is to predict the price of bitcoin 1 hour in advance with enough predictability that there is an alpha to be gained from that. And at the moment, we feel we have a very good candidate for that. And we’re looking at ways to commoditize that and obviously integrate that into bumper. So we think that’s a major step change in how the modern AI world deals with volatility and really pushes that technology to give us insights.
Baz — Bumper
I love it. Sounds super interesting. And I mean, beat the market just seems like such a big opportunity when, you know, if you’re able to know anything ahead of time. But if you can calculate that through algorithms and then use that in trading techniques, there’s just, I guess, opportunities unlocked not just for capital preservation, but for, you know, generating money as well. Let’s just, I guess, closing off, we aimed for about 30 minutes here and we’ve run to 50. So I think we’ll call the questions to a close. I wonder if I could just for fun over the last couple of minutes, is ask everybody for a prediction on maybe bitcoin and ETH price. Should we say at the end of this quarter and maybe the end of this year, 2024?
Baz — Bumper
Just be interesting to get people’s perspective on where we think the market’s going short term and longer term. So, Jonathan, you’ve been doing a lot of predictions with the AI’s. What do you know?
Jonathan — Bumper
Well, yeah, I definitely have some insight here. Yes. Reporting from the front lines, we would like to see 70 by the end of the quarter and kind of hovering just north of 100 by the end of the year.
Baz — Bumper
I can subscribe to that. I’d be happy with that. Yeah. And Moshe, from the CBR side, what are your thoughts on price predictions for bitcoin?
Moshe — CVI
Yeah, it’s actually very similar. I think that we will be a bit lower than all time high by the end of this quarter and I think that we are going to be for sure above all time high. I don’t know if 100, but somewhere between 90 to 85 by the end of the year.
Baz — Bumper
Gotcha. I can concur with both ideas. To be honest. I think that’s a lot of what the market is suggesting. With probably a little bit of help of money printing in the US, we’ll probably go a long way to helping us get there as well later this year. So we’ll see how that pans out. But yeah, you heard it here first. Not financial advice, but everybody go long.
Jonathan — Bumper
Never.
Baz — Bumper
Awesome. It’s been amazing conversation. Thank you, everybody for your time. This is recorded so we’re going to share it afterwards. But if anybody that’s listening could also click the retweet button, share the session and then we look forward to further conversations both from CVI and bumper side as time goes on. So thanks all free time. Bye bye. Great.
Moshe — CVI
Thank you, guys.
Baz — Bumper
Cheers, everyone. See you soon. Bye.
