- 2 years ago
Kaiju ETF Advisors builds, trains, and employs robust artificial intelligence (AI) and machine learning technologies designed to improve fund management decision-making. By empowering these innovative technologies to curate and provide direct management of our ETF, we’re striving to go places no Registered Investment Advisor has gone before.
Category
🗞
NewsTranscript
00:00 [upbeat music]
00:02 What is up, Zinger Nation?
00:04 Welcome back to another episode of Benzinga Live.
00:07 Happy Tuesday, everybody!
00:08 [upbeat music]
00:10 ♪ Whoa, whoa, whoa, whoa, whoa, whoa, whoa ♪
00:14 Without further ado, let's go ahead and bring my man Ryan on the stream.
00:19 Ryan, how are you doing today?
00:21 Good, Aaron. Thanks for having me back.
00:24 It's been a little while.
00:25 I know we've both been sort of crisscrossing around the globe,
00:28 and good to be back in Europe for me and on your show for you.
00:33 Yeah, it's great to have you back.
00:35 And a lot has happened since you were last on the show.
00:39 You guys ran a whole predictive AI track
00:42 at the Future Proof Wealth Management Retreat in Colorado
00:46 and recently launched a new IP fund that is valued at over $470 million.
00:52 Can you tell us a bit more about this and what this IP consists of?
00:57 Yeah, I mean, we've been talking for a while
01:00 about trying to give more direct access
01:03 to the predictive AI-based IP that we build.
01:07 And one of the things that we came up with about a year ago
01:12 was using sort of this novel Cayman Islands closed-ended fund
01:15 instead of like a regular PE offering
01:18 and inviting qualified purchasers and accredited investors into the fund.
01:24 And then that would allow them to directly buy into
01:27 the intellectual property that we build
01:29 instead of just benefit on the endpoint
01:33 from the trading systems that we implement within funds that we manage.
01:37 And yeah, I mean, because there's been this obvious awakening in AI
01:42 over the last 18 months,
01:45 the interest has been fairly substantial.
01:49 It's been kind of hard to keep up with, but I mean, I was surprised.
01:53 There were like a lot of small RIAs
01:55 that I totally didn't think would be interested
01:58 that were hunting us down at FutureProof to have these conversations.
02:02 It's pretty fascinating.
02:03 Yeah, I mean, that's amazing to see that the demand is there
02:06 even from the smaller retail,
02:08 or even from the smaller kind of wealth manager advisors.
02:13 You want to still get into AI. It's not just the big boys.
02:16 Last month, Ryan, we talked about the nature of predictability in the market.
02:20 This week, we're focusing on what AI can really do in finance.
02:25 Firstly, what benefits can AI bring
02:27 from an investment opportunity sourcing perspective?
02:30 I think it can do a lot,
02:34 but I think it's important to differentiate
02:37 between generative AI and predictive AI in this space.
02:41 So, generative AI, what can it do for you?
02:45 It can do a little bit of heavy lifting.
02:46 It can collect some information.
02:49 But in terms of delivering a high degree of certainty in any area
02:52 that you would be comfortable basing an investment decision off of,
02:55 no, it's not there.
02:56 I mean, you want to test that for yourself,
02:59 go to GPT, ask GPT to build you a robust all-weather portfolio,
03:04 and then look what it suggests.
03:05 Like, that breakdown is going to be just batshit crazy.
03:09 So, it's not going to do that for you,
03:11 but go find me all the 13 Fs or 13 Ds
03:14 that have exhibited the following criteria in the last 10 years.
03:20 It can find most of them, so that's fine.
03:23 Predictive AI, on the other hand,
03:25 which obviously I'm a little biased since that's what we specialize in,
03:28 predictive AI is very good at finding signals in the noise.
03:32 That's what it does.
03:33 It's ingesting, or at least the type that we're using,
03:37 the entire market at the tick level every single day simultaneously.
03:43 So, you've got 220 terabytes of historical data,
03:46 you have billions of daily examinations,
03:48 you have trillions of total transactions,
03:51 and it's seeing everything in real time.
03:53 So, it's going to be able to pick up on signals that we humans,
03:56 no matter how good we get or how observant we are,
03:59 we just can't get that observant.
04:02 We can't watch everything all at once,
04:04 especially relative to each other.
04:07 You're watching a sector that you're passionate about,
04:11 you think has a future, you're involved in whatever it is,
04:16 you get your little microcharts up on a 4x4 panel.
04:20 You can't watch all the prints of all of those stocks
04:24 at exactly the same time to determine where biases is appearing.
04:29 So, AI can do that, and that's powerful.
04:33 What differentiates this approach
04:36 from how a human may look to spot these trends?
04:38 And what are the challenges in training AI
04:41 to identify these market trends?
04:43 It's a really good question.
04:45 So, ultimately, what we're training the predictive AI to do
04:50 is find the result of a human investment decision.
04:54 So, if you take a step back, you think about it.
04:57 Every investment decision that's made by any investment manager,
05:01 it doesn't matter whether it's Ray Dalio or the late Charlie Munger,
05:04 or whoever, right?
05:06 You don't need to be a fly on the wall of their boardroom.
05:10 These investment management decisions will result in a trade decision.
05:15 At the end of the day, an accumulation or distribution of an equity
05:18 or combination of equities, there's a profile that's constructed.
05:23 So, you don't need to do the fundamental analysis
05:26 that went into this decision.
05:28 You simply have to find it in all of that noise.
05:32 So, that's what your reverse engineering,
05:34 these huge trillion-dollar asset managers
05:37 are not going to put these orders up lit.
05:39 And even if they put up unlit,
05:40 they're going to post a consolidated tape after 7.30 at night.
05:43 You could find them then.
05:45 So, we're not able to do that.
05:47 Humans are not able to reverse engineer that trade management decision.
05:51 So, we're trying to, what, you know, review sentiment,
05:55 try to shift bias, like, where are we seeing high-volume correlation?
06:01 Where are we seeing volume pressure at price?
06:04 And we're making our best guess,
06:07 but predictive AI is able to do this at a scale
06:09 and a level of immediacy that we can't.
06:13 And so, you know, we're not bringing our bias into this.
06:16 This is simply looking at, is the pattern there or is it not there?
06:19 It doesn't really give a shit what the actual equity does,
06:23 you know, what that equity is, what the business does,
06:26 what are its long-term, who cares?
06:28 The pattern's there, and following this pattern,
06:30 this happens most of the time, I'm going to take that down.
06:33 So, that's sort of really key to how that works.
06:36 That makes sense.
06:38 Given human intelligence is needed to train AI, Ryan,
06:41 how can we look to achieve more accurate results in these areas
06:45 that push beyond what human intelligence can achieve?
06:48 Well, I think it's important to be careful about, you know,
06:54 attributing too much capability to AI when it comes to innovation.
07:00 AI is not good at that at all.
07:03 So, if you just black box it,
07:05 you tell an AI system to "go make money,"
07:09 it will do some fairly insane things.
07:13 It will allow, for example, like a 90% drawdown
07:16 if at the end of a 10-year time horizon, it makes 35,000%.
07:21 It's like, "Yeah, but I made 35,000% in 10 years."
07:23 It's like, "Yeah, but you lost almost all of it twice."
07:27 There's no sentiment there. It doesn't really care.
07:28 It's like, "Yeah, whatever. I scored in the end."
07:31 We don't do that, right?
07:32 There's no investor that's going to sit there and say,
07:35 "Yeah, I'm cool. I'm cool.
07:36 The model says that a 90% drawdown is fine.
07:38 I'll just wait this one out."
07:40 You'd be having a heart attack, right?
07:42 And, you know, tying that in, you know,
07:46 its limitations to what, you know, one of the viewers here, Mike, asked.
07:51 It's not a standard robo-trader, and it's not intuitive.
07:55 No, it's not intuitive at all.
07:58 What these systems are able to do
08:00 is they're able to rewrite portions of their own code
08:04 relative to the criteria that they've been tasked with waiting.
08:07 So let's say you're talking about like a dip strategy.
08:11 Obviously, we have a product that does that.
08:14 And so it's looking for a specific pattern
08:17 that precedes a low-to-high mean reversion
08:20 with a high degree of certainty.
08:22 And when it sees the pattern, it takes it down.
08:24 Now, over time, the 70 different indicators
08:29 that it's looking at, the different criteria,
08:31 are weighted a certain way.
08:33 But as markets change, participants come and go,
08:36 those criteria probably aren't weighted in exactly the same way.
08:41 These over here are less correlated with positive outcomes.
08:44 These over here are more correlated.
08:46 And so it will shift within the ideology.
08:50 But, yeah, it's not intuitive.
08:52 It's not like, "This is why it's terrible at global macro," right?
08:56 You know, Russia invades Ukraine.
08:58 There's no AI that says,
08:59 "Well, here's how I think the world's going to respond to that."
09:02 No chance. Still needs us for that.
09:04 And needs us to generate the initial strategy.
09:08 It then takes it, refines it, and makes it more profitable.
09:11 That's its power.
09:13 Ryan, using Kaiju's investment products as an example,
09:17 can you provide tangible examples
09:19 of where AI has enhanced human decision-making?
09:22 Sure. So if I was to...
09:28 I'll pick a strategy that we use, or two strategies.
09:31 So we have a complementary pair of complex options strategy.
09:37 One is called BEX, the Bilateral Equity Corridor Strategy.
09:40 The other one's called RS2.
09:42 And they predominantly do different things.
09:43 So BEX is designed to trap price in between two points.
09:47 It profits off of passive premium acquisition.
09:51 RS2 profits off of massive oversold high vol conditions
09:56 reverting to the upside.
09:59 These were profitable strategies for us years and years ago
10:03 when we were manually managing hedge funds.
10:06 And we were able to achieve in the 3% per month range
10:15 in optimal conditions, which is fairly substantial.
10:18 So you're talking 36% by the end of the year.
10:23 But once we trained the machine,
10:26 the machine was able to do almost double that.
10:29 And the reason it was able to do double that was...
10:32 Well, there's two reasons for that.
10:33 Number one, it was able to weed out candidates
10:38 that we may have traded that looked good to us.
10:41 And it's looking at a deeper level.
10:44 And it's like, "No, I'm going to put that in a tier three category.
10:47 That's not going to be a winner very often."
10:49 Okay, so less mistakes, number one.
10:51 And then number two, it slightly altered
10:54 the construction of the spreads in both strategies,
10:58 the option spreads.
11:00 It skewed on the corridor strategy.
11:03 It skewed them slightly. It added a long unit.
11:05 So you give it the capability to refine
11:08 while still being true to the ideology.
11:10 At the end of the day, it didn't build a different strategy.
11:13 It still achieves what we want to do
11:15 within the risk parameters we set the way we told it.
11:18 It just sort of comes back and goes,
11:20 "Yeah, I kind of figured out a better way to do that."
11:23 And that's been shocking.
11:25 Every time I've seen the refinement, it's just shocking.
11:28 And I can't think of any way we could do that as humans.
11:32 Yeah, it's so interesting.
11:34 I mean, just figuring out where it has these use cases.
11:37 In past conversations, Ryan, you have emphasized
11:40 that AI is meant to complement human intelligence, not replace it.
11:45 What areas of the investment process do you believe
11:47 human intelligence is still required
11:49 as AI strategies start to become more common?
11:53 I think there still needs to be an initial sanity check
11:56 with respect to the ideology that you're trying to follow.
12:00 When you build a trading strategy,
12:02 you're trying to capture something that repeats.
12:05 You are trying to identify companies
12:08 whose stock is about to roll over.
12:11 If you're a short seller,
12:12 you're looking for a massive overbought condition,
12:17 which is just unsustainable,
12:18 and there's a pattern that repeats so you can short it.
12:21 Or you're buying dips, or you're trapping price,
12:24 or you're shorting or long vol, if you're an options trader,
12:28 volatility arbitrage, et cetera.
12:31 And so the ideology matches different global macro conditions.
12:37 Like trying to corridor price in the middle of the pandemic
12:41 was like a terrible idea.
12:43 It's not just going to go sideways.
12:45 That's a market condition where it's not going to go sideways.
12:48 Now, the machine, once the pandemic gets going,
12:51 realizes what market condition it is.
12:54 But as it started, we all knew how unusual this was.
12:59 You'll press the pause button.
13:00 You're like, "Whoa, maybe I'm going to sit in cash for a little bit,
13:04 and then I'll move back in."
13:06 So the machine's not great at making those decisions right now.
13:09 It's like, "Hey, opportunity, I think I'll take it down."
13:12 Completely oblivious to what's going on further out in the world.
13:16 Right.
13:17 So we still need to be the guardians there,
13:20 where we need to be more comfortable letting go.
13:23 We get asked all the time, it's like, "Oh, but you guys,
13:27 you double-check the portfolio rebalances the machine's recommending."
13:31 It's like, "You know what?
13:32 It just absorbed like 12 terabytes of data
13:39 and performed 2 billion discrete examinations in a nanosecond.
13:42 No, man, I'm not checking its math at that point.
13:45 It's freaking way better than I am at it."
13:47 Right.
13:48 And then you double-check to see if it didn't make a spelling mistake.
13:51 It's like, "Those symbols are all still there? Yeah, looks good."
13:54 But nobody looks at it and says, "I don't know, machine.
13:56 Netflix is tough. I think I'd go Disney on that one."
13:59 No way. You just got to let that go.
14:01 You're going to get outperformed.
14:03 Yeah, because you don't have the capacity
14:07 to go through all those computations and math formulas
14:11 like the AI is doing almost instantaneously.
14:14 Do you think this will cause a shift in the individuals/types
14:17 of expertise investment firms will look for
14:20 when they're hiring moving forward?
14:22 I do.
14:24 There's already been a huge shift.
14:26 I mean, if you look at sort of the granddaddy
14:30 of all this renaissance technology,
14:31 who's been doing this for 30 years, Mike, again,
14:35 asked what companies have this strategy or this technology,
14:38 I guess is broadly what he's asking now.
14:41 And you've got like only a handful of firms are using this.
14:44 Nobody's really talking about it, I guess, except for us.
14:47 The renaissance technology has like 397 scientists and five traders.
14:52 Like, that's the skew at a shop like that.
14:56 We certainly, our AI team outnumbers our traders
15:00 like three to one or four to one or something like that at this point.
15:05 You still need the trader because, you know,
15:08 once you get to a certain size,
15:10 there are counterparty conversations that need to happen.
15:13 Like, you know, the machine sort of pops up an opportunity
15:16 and says, you should trade this, you should sell this, you should buy it.
15:20 And if it's not just, you know, lit, firing out equities on the floor,
15:26 you know, you got to call a desk somewhere.
15:28 Like, hey, you know, I'm pushing 1500 of this, you know, selling the meat.
15:33 And there's a discussion that nobody's going to have with a machine.
15:36 So you need the trader.
15:38 The trader has to educate the technologists
15:41 on the mechanics of the market, how it works,
15:45 you know, what's realistic to assume, not assume,
15:47 especially when you're talking about widespreads.
15:50 So traders still have a role, but there will be fewer of them.
15:54 AI data scientists, obviously, that's the top spot.
15:58 Systematic and quant is going to go to the way of the dodo.
16:01 But I don't know any quant traders that aren't moving into AI anyway.
16:06 Like, they're super sharp people.
16:07 So they're just going to make that migration.
16:10 If they haven't already, they're going to make that migration.
16:13 And static quant funds are just-- and quant managers will just go out the window.
16:18 That makes sense. I mean, making that transition to AI.
16:22 Well, Ryan, what about risk management?
16:24 How can AI help us better mitigate investment risk?
16:27 Well, it can certainly see more faster.
16:31 And, you know, we make extensive use of that.
16:34 So, you know, our ARC system, AI risk containment system,
16:40 which is, you know, one of the more massive systems that we run,
16:44 that's essentially a holistic portfolio off-ramp mechanism.
16:49 So it's watching all the positions that we have
16:53 against their projected model trajectories.
16:57 It's watching the entire stock market at the same time.
17:02 And it's picking up minute correlations in risk across sectors,
17:07 industries, in the equity itself, et cetera.
17:10 And so it can sort of act as a bit of an early warning system,
17:15 talking at all times with our regime classification
17:18 and change detection engines.
17:20 And those three together sort of collaborate to keep us safe.
17:24 So if classification has a high degree of certainty
17:28 that this is the regime that it's in and it's happy with that,
17:32 and change detection says, "Yeah, I don't really see anything else coming,"
17:35 the ARC will look at that and say, "Yeah, this is low risk.
17:39 I don't see anything there."
17:40 But when classification says, "I don't know,"
17:43 and change detection says, "I have no idea,"
17:46 then the ARC gets vastly more aggressive
17:49 and will look to mitigate risk, minimize positions,
17:52 offload, add net long units, et cetera, make those kinds of decisions.
17:56 Understood.
17:59 Ryan, given that the AI-powered strategies have the ability
18:02 to see the whole picture when it comes to investment decision-making,
18:05 how can we on the human side of the equation
18:09 better account for our own preconceptions
18:11 and inability to see the total picture?
18:14 That's a big question.
18:16 -I know. -Big question.
18:18 How can we compensate?
18:20 I mean, we're not going to compete at a technical level,
18:25 but we don't really need to compete at a technical level.
18:28 I mean, it's useful to know the systems are there.
18:31 It's useful to know that they have predatory behavior built in.
18:37 So if the machine is not primarily
18:42 predating off of trading behavior signals that it sees,
18:46 some of the bigger systems,
18:49 I mean, I would venture say something the size of Citadels or Wellington,
18:54 something like that,
18:55 is going to be able to manage its own order flow
18:59 to trigger behaviors that it knows will follow that order flow.
19:05 -So for example-- -Interesting.
19:07 Yeah, if it's--
19:08 Okay, so you know I hate on MACD, right?
19:12 And if it's a fan-favorite retail trader by sell signal,
19:17 it's manipulated all the time
19:19 because institutions know that retail traders
19:22 will totally fire off an order on a crossover.
19:26 So if the crossover is close and it can, through its own order flow,
19:30 which it's important to note is not market manipulation,
19:33 they're actually participating in buying and selling,
19:35 that's totally legal.
19:36 If they can push order flow out that will cause that crossover
19:40 and then reverse the trade, they'll do that.
19:43 And so you need to know that that's there
19:45 and that you're being targeted on some level.
19:48 So maybe don't, you know, use five, six different criteria
19:53 to validate your entry and exits,
19:55 not just your favorite crossover signal.
19:58 That's something that's low-hanging fruit for them.
20:01 In time, I totally think that these tools
20:03 will make their way largely down to retail.
20:07 Like there's going to be sort of autonomous trade identification systems,
20:12 risk management systems, risk calculators,
20:14 that I hope, I really hope,
20:16 that the average retail trader can be like,
20:18 "You know, I'm thinking of making this trade."
20:20 You plug it in and you could get like a historical probability
20:24 that this ever would have worked or might work in the future.
20:27 That would be useful. I'd love to see those tools.
20:30 Right now, unfortunately, no one's building those.
20:32 Well, it sounds like, Ryan,
20:35 kind of the gist of our conversation today
20:37 is that use AI for the things that it's good at,
20:39 things that we humans can't do.
20:41 But at the end of the day, at least right now,
20:44 us humans are still needed to help make these final investment decisions.
20:48 And in certain cases, like a Black Swan event, COVID crash,
20:52 you're going to need humans to be able to interpret all the news
20:54 and things that are happening right now,
20:56 because models may not be built based on something that hasn't happened yet.
21:01 I mean, as always, it's been a fantastic conversation,
21:05 learning more about AI and how it will impact our investing moving forward.
21:09 Do you have any final thoughts you want to leave us with today?
21:12 I think, you know, a couple of different things.
21:16 I can quickly answer a couple of these questions here.
21:19 You know, Mike says Renaissance, a huge hedge fund using quant models.
21:23 It's actually employee owned and quant was where they started,
21:26 but they've been machine learning faced for like the last 15 years.
21:30 They in fact modeled missing commodities data
21:33 that they didn't have from the 40s.
21:34 It's kind of crazy, but you know,
21:36 nobody's getting any information out of that shop
21:38 because they also have the distinction of having the highest number
21:41 of millionaire receptionists in the world.
21:44 So there you go. Good luck getting anything out of that.
21:47 I don't think anyone would share the secrets of a 70% net of all fees shop.
21:51 In terms of access, I mean, like I said, and you've got a question here,
21:57 you've got how many individual traders use benefit from your data and models.
22:03 I wish that was something we were allowed to provide.
22:06 Sort of is one of our big complaints.
22:09 You know, regulatory restrictions right now
22:12 have just not remotely caught up to this technology.
22:15 So, you know, transparency, disseminating KPI,
22:21 saying, "Hey, this is how we do what we do.
22:22 Hey, these are models that you should consider looking at."
22:25 Not even allowed to share that if we wanted to,
22:29 because they consider it a quantitative back test.
22:31 So final thoughts,
22:33 I definitely would be careful of using generative AI,
22:37 anything other than basic heavy lifting.
22:39 I think probably most of your viewers already know that.
22:43 In terms of predictive AI, it's going to come,
22:47 but there'll be two flavors, right?
22:48 There'll be the robo-advisor style,
22:51 which I would be very, very, very careful with.
22:54 That's going to be lots of AI washing, despite the SEC's warnings.
22:58 And they'll be looking for like, "Hey, invest money with us,
23:01 and the AI will just take care of it for there."
23:03 You don't want that.
23:04 You want tools that make it easier for you to do your job.
23:07 Find me stocks that do this, you know, scanning tools, risk tools.
23:13 Show me when I should exit this based on these things.
23:16 That's what you're looking for.
23:17 In the end, I think it's reasonable to be optimistic
23:21 about what AI is going to bring.
23:23 Sadly, right now, it's a little slow on the retail side
23:26 and being a bit hoarded by the institutional side.
23:29 Yeah, I think typically with new technologies, that's how it goes, right?
23:33 The big guys get it first, then we'll wait for our access
23:36 to kind of trickle down to the retail investor.
23:39 But hey, companies like Kaiju are making it easier for retail investors
23:42 to get exposure to that world.
23:44 Well, Ryan, thank you again for hopping on.
23:46 Always great to catch up with you.
23:49 We got to do it again soon, I think in a couple of weeks, right?
23:52 Absolutely.
23:53 Always happy to be here.
23:54 And thanks to your viewers here for asking good questions.
23:58 Yeah, they came with it today. All right, good job, guys.
24:00 We will be back here in a couple of weeks with Ryan.
24:03 Ryan, enjoy the rest of your day.
Comments