Invest with AI
00:16:47 · Brett Caughran, Kirk McKeown
[00:16:47] S1: Yeah, yeah. Um yeah, and sort of, you know, like Tom Elning from Tegas always talked about, you know, selling alpha versus selling selling beta. What uh can you sort of walk us through that evolution? Because sort of felt like my rough prior was like those data sets actually were a ticket to alpha for a few years. Oh, absolutely. Like the early the early distribution was very tight. Very tight. Right. And as it broadened over time, they sort of the almost a sort of now casting dynamic happened where those data sets almost set set set an efficient first tick, sort of the day, the day they're released. Um how is how has sort of that use case evolved, do you think, of these data sets on these, specifically on these 150 tickers? Like how are funds using these data sets now versus 2012? [00:17:35] S0: Well, I think it's I think it's a great study in um alpha and how alpha moves around, right? Um, you know, alpha never goes away. It's alpha's competitive advantage, right? It's just like where does it reside and how does it get monetized? And it was a it was it's a fascinating study when you look back at it. So if you look back and and you think about sort of credit card panels, really started to proliferate in sort of the late the early teens. So call it 12, 13, 14. You know, at that time having that asset was a competitive advantage, right? And then when you turned it into a signal was a competitive advantage on top of that competitive advantage, yeah, right. [00:18:12] S1: Yeah.
00:18:12 · Kirk McKeown
[00:18:12] S0: And then, you know, sort of as it proliferated, call it 2017, 2018, you started to see setup become the competitive advantage, understanding where people were set up and all that fun stuff. And where we are today is for anybody that's trading sort of on a, I'd say, a 30 to you know 90 day setup, um, it's table stakes. You need to know what it's saying, um, but it's it's it's just another input in your process. Because it's a cartel and it's been really tight, um, the price points haven't moved as much as the alpha has. So you're you're really paying alpha prices for smart beta, um, which is which is one of the reasons CarbonArc, I formed CarbonArc. Yeah. Um I saw that I recognized that concept and realized that like something needed to change in the way data moved around. Um, because uh, you know, people were paying too much money uh for credit card data, um, that you know, sort of and crowding out all this other really cool data asset, all these other really cool data assets that if you could, if you could buy them piecemeal or you know, sort of they were at the right price, you could create a more mosaic-driven framework for understanding your names, right? Um But you know, sort of, you know, what's interesting today is what we see is um there's more vertical, just like there's vertical, like just like vertical AI is is something that we're seeing a lot of push on um in the in the in the AI movement and story, um vertical data is becoming more relevant, right? So like you know, we like we see a lot of action in TikTok data and Amazon data as people are starting to get deeper into that second derivative framework around how SKUs are moving in TikTok and how SKUs are moving in Amazon and what that means for the broader construct of you know, sort of consumer spending and also sort of brand momentum and brand heat in other parts of the market, right? And so that's a perfect example where you have your sort of market data-ish credit card panel, but then you have your sort of, you know, you like it's it's it's it's it's layering the inputs that becomes the competitive advantage.
00:20:06 · Brett Caughran, Kirk McKeown
[00:20:06] S1: Right. [00:20:07] S0: And if you think about, you know, sort of um, you know, you I know I I think you've had um, you know, sort of Rich from Arcana on, right? Like as factors have moved, you know, like what happened between 83 and 2026 in quant markets around stocks, then ETFs, then factors, um, you're gonna start seeing something similar occur in what I would call sort of alternative or external data, where you're gonna have continued uh diversification of inputs um to create sort of uh you know, sort of differentiated outcomes, right? [00:20:38] S1: Yeah. [00:20:39] S0: The alpha moves to how you consolidate your input portfolio, right? [00:20:43] S1: Yeah, yeah. [00:20:44] S0: Um I think that's a really interesting, but it has to be uh it has to be cost effective. [00:20:48] S1: Right, right. But but before we get to that, can I can I go deeper into the I think you you made a really important point on creating signal out of the underlying data? Yeah. A lot of these data firms started to basically create a research layer on top of their data asset. But a number of funds during the sort of mid-teens would ingest that raw data and actually generate superior signal. Can you just explain that concept of like the the creating the actual investable signal from the raw data asset? What are some of the the sort of processes that the best firms will do to turn that raw data into a better signal than may come out of a um one of the data vendors?
00:21:30 · Kirk McKeown
[00:21:30] S0: Well, one of the most important one of the most important things um to think about in sort of any data pipeline. And when I talk about data pipeline, I'm not talking about, you know, sort of the ETL pipeline. I'm talking about the end to end. One of the biggest challenges with data is it gets brought in by an engineer from a firm that is collecting it and continue and consolidating it on their side. So let's say it's a credit card panel before they're putting their own signal on top of it, which is always um a questionable sort of thing from a For a lot of reasons. You get the data in, it's brought in by an engineer. That engineer brings it in, does something to it, and then your scientists will do something else to it to clean it, dedupe it, make it to the get it, tag it, all of the things. Then it goes into statistical models, and it's all historically been linear regressions, short time horizons, and a lot of bias embedded across all of that, right? So before it even gets to the decision maker, you know, the PM and the analyst, I'm talking about the discretionary side, not the quant side. That's a different, whole different conversation, ball of wax. You know, it's gone through three or four pairs of hands, uh uh uh tens of hours of processing, and you know, they have to make a risk decision. You still have to pick a stock. At the end of the day, you still have to pick a stock. So you got all of that. And what like and the thing that you know, sort of I've learned through the years, and I've been doing this for a little while, is you always have to make sure that the data marries to the business that you care about, right? So, you know, sort of um a good example of that is looking at like dollar stores or looking at like discount, like ollies. Ollie's, which I don't even know if it exists anymore, it's just always is stuck in my head as a good example. It was 50% cash pay or something like that, right? So understanding cash pay versus you know, sort of all these things, you have to figure out what payment structure works for the company to make sure it ties back to the data, right? And and then is it so the first order isn't whether or not it correlates, it's whether or not it makes sense to use it as an input in your process. Right. And if it does make sense, great.
00:23:34 · Kirk McKeown
And if it doesn't, because of something like a cash pay mix, it might be a good read of the consumer or the low-end consumer anyway. So you might use it, but again, it comes back to waiting. These are mosaic decisions, semi-systematic decisions that require sort of like a massive amount of domain expertise to go in. And that's what I think the largest places have figured out with their businesses, which is the more you can inject into the into the data structure and into how you're thinking about your model, geographic bias, demographic bias, um, key drivers in the business, the second derivative of segment margins. So think about it, like tractor supply, right? I've used this example a few times. Tractor supply, I think 20% of their stores are in Texas and 20% of their business is animal feed. If weather in Texas is 2x colder than usual, means the cows and horses are eating more because they eat three times their weight in the in the cold. So you could see a comp boost in cold weather markets for you know, sort of animal feed, but it's it's like it's like it's like flu traffic for uh hospitals. You know, it's it's it's good traffic, bad margin, right? And so like knowing it's like lumber at at Home Depot, it's a comp boost, but not a margin boost, right?
00:24:42 · Brett Caughran, Kirk McKeown
[00:24:42] S1: Like I covered managed care, like you'd see the sort of a trend spike, but then it wouldn't show up in Q1 MLR because the sort of claims from the providers are on a 60 to 90 day leg, so it doesn't show up until Q2. And so there's all these sort of great nuances in all of these businesses. Like you actually sort of need to understand the business model to like, okay, we want to actually leg this by a quarter because there's actual uh a sort of a causal logic behind doing that. [00:25:08] S0: And and and and getting that right inside of a large organization is really tricky and requires a ton of collaboration and trust throughout the org, right? Because, you know, sort of especially at the big shops, you know, sort of um, you know, it and so it's important to embed as much of that into the centralized research function as possible because it's a bona fide's credibility thing, right? It's like you show up and talk like that with a PM, they feel more comfort taking data from you, taking research from you because they're like, okay, might not get it right all the time, but they're thinking about it the right way. [00:25:40] S1: Yeah. And then you talked about setups, and they're like, we're we're two hedge fund guys, so we know what setup setups mean, but maybe not all of our viewers uh viewers understand that. Like what what's a setup?