The One Factor That Explains the Struggles of Value, International and Small-Cap Stocks | Kai Wu

[00:00:00] Value investing cannot be dead. The idea of buying undervalued assets is, by definition, what should lead to future returns. The challenge is more how you measure it and what you're measuring. If you're an investor who focuses only on that first half of the equation, just the term that is the tangible, and ignore intangible assets, to the extent that intangible assets are priced, that there is some value there, you're going to be systematically short in underweight, innovative companies, firms with modern business models. It's not about US versus international. It's about intangible versus tangible.

[00:00:29] What's happened is that the intangible economy has taken off, and the tangible economy has stagnated. One really interesting thing that I found when I was doing this research is that the correlation between intangible value and traditional value is basically zero.

[00:00:43] Welcome to Excess Returns, where we focus on what works over the long term in the markets. Join us as we talk about the strategies and tactics that can help you become a better long term investor.

[00:00:53] Jack Forehand is a principal at Validia Capital Management. Justin Carbonneau is a managing director at Life and Liberty Indexes.

[00:00:59] No information on this podcast should be construed as investment advice. Securities discussed in the podcast may be holdings of clients of Validia Capital.

[00:01:05] Hey guys, this is Justin. In this episode of Excess Returns, Jack and I are joined by Kai Wook, founder of Sparkline Capital. Kai shares his insights on intangible value investing, which leverages techniques like machine learning and natural language processing to identify firms rich in intellectual property, brand equity, human capital, and network effects.

[00:01:19] Kai walks us through why US firms have outpaced many global counterparts by investing heavily in intangible assets.

[00:01:24] While traditional industries abroad have stagnated, but Kai's findings reveal an interesting opportunity, a subset of international firms with strong intangible value trading and attractive valuations.

[00:01:33] This conversation highlights the evolving role of intangibles in value investing and the potential for international diversification without sacrificing innovation and returns.

[00:01:40] As always, thank you for listening. Please enjoy this discussion with Sparkline Capital's Kai Wook.

[00:01:43] Hey Kai, welcome back. Thank you for joining us.

[00:01:46] Great to be back.

[00:01:48] We haven't had you on in a little while, and I just want to say congrats on the new additions in your life, including the new International Intangible Value Fund.

[00:01:59] Thank you.

[00:02:33] Thank you for listening.

[00:02:59] And a lot of other things that I think investors can learn a lot from.

[00:03:02] So highly recommend you going there and subscribing to his research.

[00:03:06] Thanks, Justin.

[00:03:09] Yeah, you're welcome.

[00:03:10] So to start, let's just, you know, before we get into that stuff, let's just talk about a little bit the struggles of, I guess, traditional systematic value in the past decade.

[00:03:20] And we wanted to just ask you your thoughts on, do you think it's mostly because the way that investors are looking at these companies are, they're not really looking at the right things when it comes to trying to uncover what the value is here?

[00:03:36] Yeah, I'd say that's probably right.

[00:03:39] Right. So, you know, as you're alluding to Justin, just to set the table a little bit.

[00:03:43] Right. So like the Fama French value factor spent, you know, I think the data goes back to 1926.

[00:03:49] Right. The first 80 years consistent of performance, the past 15 or so, not so much.

[00:03:54] Right. And so the question, of course, is, you know, is value investing dead?

[00:03:58] Is this broken? Are we never going to see returns ever again?

[00:04:01] And, you know, what you're alluding to Justin, I think, is the key insight here, which is, you know, of course, value investing cannot be dead.

[00:04:09] Right. The idea of buying undervalued assets is, you know, by definition, what should lead to future returns.

[00:04:13] The challenge is more how you measure it and what you're measuring.

[00:04:16] And so to go back to Fama French and the value factor, as most people define it, using kind of tangible book value, that the challenge there is, you know, book value is a component of the intrinsic value of a company,

[00:04:29] but just one of a few different things. You know, I like to think of it as basically being intrinsic value equals book value plus or tangible value plus intangible value.

[00:04:38] And if you go back to, you know, the 1930s when Ben Graham wrote security analysis, you know, which was, you know, in industrial time, that made sense.

[00:04:47] Right. That was pretty much the entire game.

[00:04:50] We fast forward to today and, you know, Apple and NVIDIA, Google, it's really the intangible assets, the intangibles out of the balance sheet that matter.

[00:04:56] And so going to your point, if you're an investor who focuses only on that first half of the equation, right, just the term that is the tangible and ignore intangible assets,

[00:05:06] to the extent that intangible assets are priced, that there is some value there, you're going to be systematically short and underweight innovative companies, firms with modern business models.

[00:05:16] Right. If you don't give Google credit for its IP and network effects, you're going to be always underweight Google, which has not been a good trade.

[00:05:23] And so I think you can, you know, pretty intuitively understand that a lot of the struggles of value the past decade or longer has been to this mismeasurement problem,

[00:05:33] where you are kind of ignoring intangibles and as a result, always underweighting the things that have gone up.

[00:05:39] And in my opinion, at least likely will continue to do OK, you know, as we move forward.

[00:05:44] What are your thoughts on should or could intangible assets be incorporated in standard accounting data or metrics?

[00:05:56] And, you know, depending on where you fall, what would be maybe the downside if it could be?

[00:06:03] Yeah. So look, I think so the challenge here with, you know, at least in the US, right, with GAAP accounting is that financial statements don't take take an inconsistent approach with how they treat physical and intangible capital expenditures.

[00:06:19] So physical CapEx, if I spend money building a factory, gets capitalized, it goes on your balance sheet and it gets depreciated, whereas intangible assets are expensed.

[00:06:29] So they come share to the top line, which is effectively punishing firms that invest in intangibles as opposed to physical businesses.

[00:06:36] And so one thing that's been proposed and, you know, within the academic literature is, you know, now I think relatively accepted is this idea of just treating these two things consistently.

[00:06:46] So let's treat intangible investments, like R&D, the same way we treat physical CapEx.

[00:06:53] And if we do that, it helps.

[00:06:55] You know, I've actually run this study.

[00:06:56] So I said, let's take the Fama-French factor, which looks at book value, and then to book value add this additional component of capitalized intangibles defined as being, you know, R&D and marketing and things like that.

[00:07:09] What you find is that it helps, but is no panacea, right?

[00:07:12] So instead of being in a, you know, 67% drawdown, you're in a 65% drawdown, right?

[00:07:18] So like it's, you know, theoretically correct and is potentially helpful, but in practice, it's not making a huge difference.

[00:07:25] And so why would that be?

[00:07:26] I think this is an important question to ask.

[00:07:29] And, you know, you think about why the folks who have been against, why did the accountants not want to capitalize intangibles in the first place?

[00:07:36] The reason is because the relationship between the money invested today and what you get out is pretty weak, right?

[00:07:43] There's two companies that spend, you know, $50 million on trying to develop a cancer growth, right?

[00:07:49] One company, it ends up failing trials and is worth zero.

[00:07:53] Company two ends up being a billion-dollar blockbuster, right?

[00:07:56] I think this phenomenon of these kind of like nonlinear effects happens across R&D and viral marketing, even in, you know, hiring of talent.

[00:08:05] And so in as much as that might be the case, anything that's historical cost-based, which by definition accounting is, is not going to work, right?

[00:08:13] That's like saying a dollar investment from Boeing and from SpaceX is the same, which it has not been, obviously.

[00:08:17] And so we're much better off trying, at least conceptually, to focus less on the money spent in versus like the output created as a result of sentiment.

[00:08:26] Do you think the money spent in is less predictive in the intangible world than the tangible world?

[00:08:30] So, for example, like if I build a plant, it may work out, it may not work out.

[00:08:33] So there's variability there too.

[00:08:35] Is it more variable like in the intangible world?

[00:08:37] Yes.

[00:08:38] I think that's the common, I think the common understanding is that the answer is yes, that there is a lot more kind of unpredictability in the results.

[00:08:45] So how do you think about measuring?

[00:08:47] So obviously you don't think the accounting is enough to measure intangibles.

[00:08:50] So can you sit at a high level, talk about how you value intangibles?

[00:08:54] Yeah.

[00:08:54] So look, accounting information is insufficient, but look, the good news is that 99% of information is not accounting information, right?

[00:09:03] And we happen to live in 2024.

[00:09:05] We've seen an explosion of, you know, big data and structured data, information, you know, created through individuals and all these interesting alternative data sources.

[00:09:14] So what I did was I said, let's like look through all this information and figure out are the things that XCANSI might be potentially informative with regards to intangible assets.

[00:09:23] Examples might include patent abstracts or LinkedIn, Glassdoor, any sort of like information like that.

[00:09:30] That's not necessarily in a spreadsheet, it's not a PE ratio, but could potentially have information on intangibles.

[00:09:36] And that, you know, is actually really interesting as a potential, you know, well-drawn.

[00:09:42] The challenge has been historically, right, that the tools haven't really existed to do this.

[00:09:47] So, you know, go back 10 years even.

[00:09:49] At that point, what are you going to do?

[00:09:51] Take a patent abstract and throw it through linear regression?

[00:09:54] It's not going to give you any interesting information.

[00:09:56] So in order to parse this information and to create factors around it, you need new tools.

[00:10:00] And we've been very lucky, right, that we've also, you know, obviously been through this kind of renaissance in natural language processing tools, large language models and AI, ChatGPT, which are enabling us to take this information and structure it.

[00:10:15] Right?

[00:10:15] I can take a, you know, we were previously about to take, you know, a million patents and read through each abstract one by one, just not scalable, obviously, to figure out, say, which ones are related to AI.

[00:10:25] Now I can just throw that through ChatGPT, you know, loop over using the API, and you end up with, you know, a score, a structured score that has been derived now from the unstructured text.

[00:10:36] And so I think there's a lot of potential in this sort of use case to use this alternative data, this unstructured data, as ways to quantify intangible assets.

[00:10:48] Yeah, it's funny.

[00:10:49] I was thinking, like, all of us are getting all pumped up about, like, AI and all this stuff recently.

[00:10:52] I'm like, Kai's been doing this for years.

[00:10:53] Like, you've been doing this for, like, a long time, but all of us are just getting all excited about it now.

[00:10:57] Yeah, the funny thing is that, you know, we were trading large language models, like, in 2019.

[00:11:02] Right?

[00:11:02] This is, like, the BERT model, like, the original transformer that Google came out with.

[00:11:07] It was open source.

[00:11:08] And so we were trading it ourselves.

[00:11:10] You know, nowadays, you would never do that, right?

[00:11:11] Unless you have, you know, a trillion dollars and, you know, thousands of GPUs, it doesn't make any sense.

[00:11:17] And so I think, in a way, it's become, you know, both good and bad for us.

[00:11:22] You know, on the plus side, we used to talk about doing this stuff and people would look at us like, well, crazy.

[00:11:27] Right?

[00:11:28] What are you talking about?

[00:11:28] These, like, you know, AI models?

[00:11:30] That's crazy.

[00:11:31] Nowadays, everyone recognizes what it is because they know ChatGPT.

[00:11:34] But on the other hand, the tools are now more democratized, right?

[00:11:37] Which is, like, anyone with an open AI account can use their API and it will be better than what anyone can do in-house, aside from maybe five companies in the world.

[00:11:47] But, you know, I'd say in general, it's definitely been positive.

[00:11:50] Yeah, I would think so because people now understand what you're doing.

[00:11:52] Like, you know, probably when you first started doing this, people were like, what is this voodoo magic you're doing here to come up with these intangible values?

[00:11:58] And now they're like, I totally understand what this is because I see it in front of me.

[00:12:01] Right.

[00:12:01] Exactly.

[00:12:02] Yeah.

[00:12:03] So you have four pillars, I believe, when you look at intangible value.

[00:12:07] Can you talk about what they are?

[00:12:08] Yeah.

[00:12:08] So when we first set out to, you know, quantify intangible assets, what we wanted was a taxonomy that, in a parsimonious way, explained, you know, as many different types of intangibles that are out there.

[00:12:19] And what we settled upon was a four-pillar framework.

[00:12:22] So the first one is intellectual property or innovation.

[00:12:25] The second one is brand equity.

[00:12:27] Third is human capital.

[00:12:28] And then finally, network effects.

[00:12:30] And the idea is that, you know, through these four categories, we can more or less capture the universe of assets that might matter in an intangible sense.

[00:12:38] I'm just wondering, like, as this technology advances, like, it seems like it has advanced a ton, like, in the past year.

[00:12:43] Does this allow you to value intangibles better as it advances?

[00:12:47] Or have you kind of just been using the techniques before all of us, the rest of us, became aware of them?

[00:12:53] Yeah.

[00:12:53] I think it makes, yes.

[00:12:57] As the tools get better, think about it this way.

[00:12:59] Like, there's a lot of value in this unstructured data that's untapped.

[00:13:03] And, you know, when we first set out to do this, it was like word banks.

[00:13:07] You may have read, like, the Laughlin and Donald paper, right, where they find a keyword search.

[00:13:11] That's kind of the original attempt to create factors just by counting words.

[00:13:15] And then came embeddings, which are like single-layer neural networks.

[00:13:18] And then came large language models.

[00:13:20] And with each step, we would capture a greater and greater share of the insight in these documents.

[00:13:25] And so we're not at 100%, obviously, but we're, you know, doing much better than we used to.

[00:13:30] You know, just from the standpoint of when you're doing keyword search, the problem is that if you're saying,

[00:13:35] this company is not innovative, right, it's not going to take up the word not.

[00:13:38] Embeddings help a little bit, but, you know, large language models are really capturing the most nuance and riches in language.

[00:13:44] And, you know, even within there, obviously, it's not, you know, binary.

[00:13:48] There's, of course, there's GPT-1, 2, 3, 4, so on and so forth.

[00:13:51] As the tools get better, the way I see things trending is we're going to be able to capture a greater and greater share of the information currently kind of embedded in this unstructured data.

[00:14:03] Kai, with those four pillars, could you just give us a quick sort of example in each one of, like, what you would look at?

[00:14:12] And how you would look at it in the data?

[00:14:15] Yeah, so, you know, I mentioned the patents already for IP.

[00:14:18] That's kind of the most straightforward example.

[00:14:20] You want to look for companies with lots of patents, but more importantly, they need to be kind of leading edge, not lagging edge, right?

[00:14:27] Like, if I patent today on some kind of, like, obscure stand that doesn't matter or some obsolete technology which I wouldn't even get the patent,

[00:14:33] that's not that useful.

[00:14:35] You know, human capital is another good example.

[00:14:38] You know, one thing we look at is, you know, LinkedIn.

[00:14:41] And so with LinkedIn profiles, you can basically see at each point in time what a company's entire workforce looks like.

[00:14:47] You know, what sorts of people do they hire?

[00:14:49] What sorts of talent do they obtain?

[00:14:51] All else equal, you want companies that are hiring top talent, folks with, you know, innovative experiences, good degrees,

[00:14:58] who have worked at, you know, competitors and kind of prestigious companies.

[00:15:03] One thing we built was a page round algorithm, kind of the backlink idea,

[00:15:07] where we said companies that hire from other good companies, right, are better.

[00:15:11] And kind of like, when you trace it all back, it leads to a signal that, you know, it's basically like a recruiting score.

[00:15:17] Companies that are able to pull talent are, you know, probably, that probably proxies for something pretty good.

[00:15:22] You know, culture, we look at Glassdoor as another example.

[00:15:25] You know, one thing we recently added was trademarks.

[00:15:29] So the USPTO, the US Patent and Trademark Office, is the, you know, organization that, with which you file patents,

[00:15:36] which we've been looking at for a long time.

[00:15:39] You know, one thing we asked was, interestingly, within the financial literature, a lot of, not a lot,

[00:15:43] some people are looking at patents.

[00:15:45] It's not uncommon things for people to look at.

[00:15:47] But almost nobody, as far as I could tell, was looking at trademarks.

[00:15:50] Yet it's kind of a sister, right, like sister concept to patents.

[00:15:54] So we went out and actually built up the database by scraping the USPTO website.

[00:16:01] And one of the funny things, actually, was, you know, when I was doing that, like, I noticed there was, like, a corrupt file.

[00:16:06] So I actually emailed the government and said, hey, guys, like, you know, this file's messed up.

[00:16:10] Like, what's going on?

[00:16:12] No response.

[00:16:13] I ended up just figuring out a way around it, so it wasn't a problem.

[00:16:16] And then, like, three months later, I got an email from, like, somebody who was there and said, oh, wow, like, I didn't even notice.

[00:16:21] No one's asked about that.

[00:16:22] It's so weird.

[00:16:22] But, yeah, it's fixed now.

[00:16:23] Which, to me, suggests that, like, yeah, this is really definitely under the radar of the data set.

[00:16:27] No one's really looking at it.

[00:16:28] But it's pretty useful, right?

[00:16:30] Like, when we talked about this, when we did this research, there was around the time that that Barbie movie had come out.

[00:16:35] And so, you know, people were talking about Mattel and Barbie.

[00:16:37] And we said, you know, the IP, this brand concept of Barbie, like, we can trace its origin from, like, being a doll to, like, all the accessories, to the movie, to the website, so on and so forth.

[00:16:46] And kind of see the, you know, universe being created around it.

[00:16:50] And, you know, it turns out that similar depends companies that, you know, hold a lot of valuable trademarks, that's an undervalued intangible asset.

[00:16:59] That, you know, folks, that most investors are not truly giving companies with these kind of strong brand capital enough credit.

[00:17:06] And their stock have continued to outperform subsequently, you know, relative to the market.

[00:17:13] So once you've gone through the four pillars and you've got your values, how do you think about turning that into a value strategy?

[00:17:18] Yeah, so this is where it gets boring.

[00:17:21] So I was at GMO for a long time.

[00:17:22] That was where I started working.

[00:17:24] They're a pioneer in quant investing from the 70s and 80s.

[00:17:27] And, you know, what I was taught was always, you know, to prioritize transparency and, like, intuition over optimality, right?

[00:17:34] So you get a more robust signal.

[00:17:36] And so rather than throw all these signals, so now imagine I have dozens of signals into an optimizer and just let it run.

[00:17:42] That's not what I'm doing.

[00:17:43] So instead I'm saying, let's come up with a very simple, like, transparent way of going from point A to point B.

[00:17:48] What I do is I create the four pillars and I say, take all the signals that are tied to one pillar and then average them within that pillar.

[00:17:55] So, for example, to our AI thing, you know, you can quantify AI by looking at job postings.

[00:18:01] So which companies are hiring employees or looking to hire employees who are what you would consider AI human capital.

[00:18:09] Folks who, you know, know, like, CUDA or, like, TensorFlow.

[00:18:13] You can also look at the patents.

[00:18:14] Which companies are patents in the AI?

[00:18:15] So on and so forth.

[00:18:17] And those things are going to be correlated.

[00:18:18] And so what you don't want to do is say, if I have 99 AI signals and one brand signal, then I'm just going to equal weight them, right?

[00:18:25] Because then you end up just, this first thing dominates.

[00:18:27] So instead what you do to deal with the correlations is to just average them first at the pillar level before you sum.

[00:18:33] And that kind of takes care of a lot of this while still kind of has the added benefit of giving you this intuition where I can say, hey, this company is something we like because of its strong brand.

[00:18:43] Whereas this is something we like more because of its human capital.

[00:18:45] And so the final score, the composite score, just ends up being kind of the blend of the four pillar scores.

[00:18:51] And so you have, you know, for N companies, you have N scores.

[00:18:55] So it'd be something similar to what those of us in the value world would think of as like a value composite.

[00:19:00] Exactly.

[00:19:01] Right.

[00:19:01] So it would be similar to saying I'm more going to kind of blend across EBITDA, price to book, price to earnings, price to sales.

[00:19:08] You wrote a great paper where you looked at the performance.

[00:19:10] You sort of tested this in the same way academics would test a factor.

[00:19:13] Like, can you talk about how it performed and how it looked against the other factors?

[00:19:17] Yeah.

[00:19:18] So, yeah.

[00:19:20] So what, so my starting point was as a baseline was to say, let's look at the Fama-French construction, right?

[00:19:24] So let's imagine, I mean, they have different specifications, but imagine we're looking at the top, you know, quintile of value on first the book versus the bottom, right?

[00:19:34] And as I mentioned before, it used to work well, has not worked the past 15 years.

[00:19:38] And so I said, what if we use the exact same construction long short, but use as intangible value scores instead of the price of books?

[00:19:44] And what you find is that, you know, for the first, I started in like 95 was the beginning of the backtest.

[00:19:50] And for the first 10 years or so, it didn't help that much.

[00:19:53] It was like somewhat useful.

[00:19:54] But where it really shined was after the financial crisis, when the traditional value factor had this huge drawdown.

[00:20:00] That's when the intangible value scores factor continued to perform, you know, right along trend.

[00:20:06] So it was like no noticeable, you know, dip.

[00:20:09] And I think that speaks to the fact that, you know, over time it has been the case, right?

[00:20:13] As we kind of started the conversation discussing that the economy has transformed and that the companies that are minors today are intangible companies, less industrial companies.

[00:20:22] And so as that becomes more and more important, you know, the traditional value factor has suffered, whereas this thing has continued to do okay.

[00:20:30] How far can you get back data on the intangible value factor?

[00:20:33] So it depends on the data source.

[00:20:35] I mean, patents go back to 1790.

[00:20:37] So if you wanted to get back to a patent factor, your limiting thing would be the market cap, the numerator, not the denominator, actually.

[00:20:45] Obviously, you know, LinkedIn and, you know, I mentioned Glassdoor, any social media feed, that's going to go back less far.

[00:20:51] You know, what we've done with the 95 example, we started with 95 because that was when the Capital IQ database kind of picks up coverage.

[00:20:59] And starting with that, we kind of add data sources as they become available.

[00:21:04] And so, you know, that's kind of been the approach so far.

[00:21:07] How do you think about this relative to traditional value?

[00:21:09] Because I can make two arguments.

[00:21:10] I can make one argument.

[00:21:11] This is just a better way to measure it.

[00:21:12] So this would be a replacement for traditional value.

[00:21:14] And then the other argument I can make is traditional value has been out of favor forever.

[00:21:17] It's probably going to come back at some point.

[00:21:19] Maybe I should blend the two together in some way.

[00:21:21] Like, how do you think about that?

[00:21:23] Yeah, I mean, that's been an approach that I've advocated as well.

[00:21:25] So one really interesting, like, thing that I found when I was doing this research is that the correlation between intangible value and traditional value is basically zero.

[00:21:35] Like, it fluctuates through time.

[00:21:36] In the late 90s, early 2000s, they were both correlated, positively correlated because they were both underweight, you know, the tech bubble.

[00:21:43] Then they were negatively correlated.

[00:21:44] Now they're more like around zero.

[00:21:46] But again, over the full arc of time, uncorrelated.

[00:21:50] And it seems maybe a little counterintuitive because in theory, they're doing the same thing.

[00:21:53] They're both looking for price X, right?

[00:21:56] They're looking for cheap price intangible to intrinsic value assets.

[00:21:59] But I think if you think about it more, the difference is this, which is go back to the equation I first started with, which is intrinsic value equals tangible value plus intangible value, right?

[00:22:08] What is traditional value doing is it's focusing on finding assets that are cheap in the tangible side of the balance sheet, whereas the intangible assets is looking at the intangible side of the balance sheet.

[00:22:17] And the types of names that come out are very different, right?

[00:22:20] Traditional value usually gets kind of like, you know, gets made fun of because it owns, you know, banks and energy companies, materials, kind of boring old economy names.

[00:22:30] Intangible value is the opposite.

[00:22:31] It holds tech companies, consumer, you know, discretionary services businesses.

[00:22:37] And so there's very little overlap both in the names and the sectors.

[00:22:40] So it kind of, you know, makes sense that they'd be the complementary rather than kind of substitutes in a portfolio context.

[00:22:46] So on the idea of this being complementary to the other factors, you actually developed like a six-factor model in one of your papers where you looked at adding this to the other factors.

[00:22:53] Can you talk about what it did when you looked at it on a multi-factor basis?

[00:22:56] Yeah.

[00:22:57] So look, this paper was obviously targeted towards a very specific audience, those of us who are considered factor investors and build multi-factor portfolios.

[00:23:05] The idea here was, you know, start with a baseline five-factor portfolio.

[00:23:09] That's market, small cap, quality, momentum, value, I believe.

[00:23:14] And then add a sixth factor, this intangible value factor.

[00:23:18] And the first thing I looked at before even looking at any, like, backtests was just, like, what is the kind of correlation behavior across these factors?

[00:23:25] I already mentioned that the correlation between traditional value and intangible value is close to zero.

[00:23:30] It's also close to zero with size and momentum, which is just a fast burn signal, and market.

[00:23:36] What was kind of interesting was actually the quality factor, right?

[00:23:39] Because intuitively, quality is basically stocks that have, you know, economic moats that allow it to earn high profits.

[00:23:46] And what is a moat but, like, a patent?

[00:23:49] A patent's a moat, so is a luxury brand.

[00:23:51] So intangible assets are part of the modern moats.

[00:23:54] So it's a little bit unintuitive that they'd be zero correlation.

[00:23:57] Like, why would that be the case?

[00:23:59] I mean, so I actually did a study in this paper, which I think was, you know, pretty interesting, where I said, let's look at stocks with high intangible value.

[00:24:07] What does the ROE, the return on equity, look like this year as well as next year, the year after that, all the way through 10 years later?

[00:24:15] And what you find is that actually companies that are investing heavily in intangible assets have lower profitability this year because they're investing up front to get the back-ended payout, right?

[00:24:24] You have to put money into getting the patents, but you can't actually monetize it until you get it.

[00:24:28] But then as you kind of go forward, there's a J curve, and it starts to be really profitable.

[00:24:33] And so the insight there is that, you know, what is intangible value?

[00:24:37] It's like the quality of tomorrow.

[00:24:39] There's an inter-temperal disconnect between the same thing, just, you know, once in the future, once today.

[00:24:45] You know, traditional qualities looking at, like, trailing profitability or next year consensus profitability, whereas intangible value is looking at profitability 10 years down the line, standing from the investments made today.

[00:24:56] Okay. So anyways, to go back to the main thread, the point is here is that intangible value turns out to be uncorrelated with all the other five factors.

[00:25:06] Therefore, adding it to a five-factor portfolio, if you have such uncorrelated things that have positive expected returns, you get a better outcome.

[00:25:13] And that's what we find.

[00:25:15] Yeah, it's interesting. I was thinking about Google when you were talking about that, because, like, all these MAG7 firms right now are just spending outrageous amounts of money on AI.

[00:25:22] So probably the phenomenon you talked about is probably what's happening, right?

[00:25:24] You're seeing maybe a decline in some of that ROE stuff, and you'll probably see some big boost in the future from it.

[00:25:30] That's the hope, yeah.

[00:25:31] I mean, and that's the history of these firms.

[00:25:32] Like, take Amazon. Amazon's famous for years and years and years.

[00:25:36] They never even had positive earnings, right?

[00:25:38] At least gap earnings.

[00:25:39] But, like, we all knew what they were doing, right?

[00:25:41] They were reinvesting in the business.

[00:25:43] They were, you know, pricing their services very low to earn consumer goodwill to build network effects to eventually lock in the retail market.

[00:25:49] They were investing in what eventually became AWS, the cloud business, which is, you know, a profit center.

[00:25:55] And so that's the thing.

[00:25:56] You have to spend money to make money.

[00:25:58] And, you know, traditional value and traditional quality even approaches in this case would generally discount a firm like Amazon saying, hey, look, they're just, like, not that profitable.

[00:26:07] They're, like, you know, whatever.

[00:26:09] But, like, you know, these are the companies that ultimately, you know, ended up becoming the best investments.

[00:26:14] And so looking for firms, the next Amazon, so to speak, that they're investing today for the next 10 years.

[00:26:20] That's kind of the idea here.

[00:26:22] How do you think about turning this into a portfolio?

[00:26:24] So, you know, a lot of quants might just take their highest scoring X number of stocks, equal weight them, and that becomes the portfolio.

[00:26:29] How do you differ from that?

[00:26:30] Yeah, I mean, look, that's a totally viable approach.

[00:26:33] I think, you know, as of one, like, you want to maintain flexibility as the best way to put together a portfolio.

[00:26:38] You know, I think I mentioned earlier in the context of the form of French, you can look at long short factors.

[00:26:44] You can look at a long only portfolio that, you know, over and underweight stocks relative to a market.

[00:26:48] You could apply, in theory, like sector neutralization and other constraints.

[00:26:52] My preferred approach, to be honest, is to not do that.

[00:26:55] But I think that, you know, sector definitions are kind of arbitrary, right?

[00:27:00] Like, Amazon, is it a tech company or is it a retailer?

[00:27:03] You know, is Tesla an automaker or is it a software company?

[00:27:07] And so there's some arbitrariness to these definitions that I think really constrain you.

[00:27:12] And you're almost better off looking bottoms up and saying, which companies are attracting the best talent?

[00:27:17] Which companies are trading the most IP?

[00:27:19] Regardless of sector.

[00:27:21] And if it happens to point you to technology today, so be it.

[00:27:24] And if there's a, you know, a nuclear renaissance or a green revolution and, you know, the biggest companies in the U.S. are energy companies in 10 years, so be it.

[00:27:33] Like, I think you want to maintain flexibility to do that.

[00:27:36] And I think especially in a long only context, that's, you know, the way forward.

[00:27:39] I think in long short, if you're highly levered, you want to be, you know, a bit more careful about your risk model.

[00:27:43] And that's the world I come from, the hedge fund world.

[00:27:45] But at least, you know, in a long only context, I think that the kind of unconstrained approach is probably superior.

[00:27:52] I want to ship to talk about international stocks because you just wrote a really good paper about this.

[00:27:56] And I want to kind of ask the same question, Justin, at the beginning, but in a different way.

[00:28:00] Like, Justin asked, is the underperformance of value investing explained by intangibles?

[00:28:04] I want to ask you, is the same true here in international stocks?

[00:28:09] Yes. So, basically, in this paper, what we did was we said, let's expand what we've done in the U.S.

[00:28:17] So keep in mind, like, I wrote this paper you're mentioning, I wrote this year, which is about three years after the work I did in the U.S.

[00:28:24] And the idea was to start in the U.S. and then ultimately eventually expand and test on a kind of quasi-out-of-sample basis to see how this works abroad.

[00:28:33] And so I basically replicated the entire, like, model globally to the extent possible.

[00:28:39] And, you know, so instead of having 3,000 stocks in the U.S., we have now, like, 9,000 stocks or whatever, you know, and we'll develop an EM.

[00:28:47] And, you know, so basically built out the same factors here.

[00:28:49] And a lot of the results are the same.

[00:28:52] But, yeah, you know, happy to talk about anything related to the paper.

[00:28:56] Yeah. So international stocks in general are much less exposed to technology, right?

[00:29:00] Like, the weighting is much lower in technology.

[00:29:02] I mean, is that a big explanation as to why they underperformed or does it go deeper than that?

[00:29:07] Yeah. So if you think about underperformance, I think, you know, the first thing I do in this paper is actually probably the most important and has nothing to do with intangible assets,

[00:29:13] which is I do a decomposition of returns.

[00:29:17] And so here's a setup, right, which is over the past 13 or 14 years.

[00:29:21] So around the same time that value started underperforming in the U.S., international stocks started lagging the U.S. stocks, you know.

[00:29:29] And I think the gap is pretty wide.

[00:29:31] I think it was about 8 percentage points per year, which compounds to obviously a huge difference, right?

[00:29:37] Effectively, international stocks have been dead money the past decade, whereas U.S. stocks have just died.

[00:29:41] And the question is, you know, what has been the driver of that?

[00:29:44] Has it been valuation changes, as some value investors might argue, or has it been fundamentals, right?

[00:29:49] Because in theory, if, you know, U.S. stocks outperformed international stocks by 8% per year in terms of earnings growth,

[00:29:56] then there's nothing to talk about here.

[00:29:57] And they just grew better.

[00:29:58] So the answer is somewhere in between, where it turns out that about 5% or 5.5%, I think,

[00:30:03] was a gap between international and U.S. annual EPS growth in U.S. dollar terms.

[00:30:09] In absolute numbers, U.S. stocks grew up 5.5% per year, which is pretty good.

[00:30:15] And emerging market stocks, sorry, international stocks had negative 10 basis points.

[00:30:20] Yeah, real earnings growth per share in U.S. dollar terms.

[00:30:23] So in other words, then they were stagnant for the past decade.

[00:30:26] And that explains like two-thirds or so of the, you know, 8 percentage point gap.

[00:30:31] The other, you know, two-and-a-half points difference comes from changes in valuation ratios,

[00:30:37] where the international world basically had a flat fee ratio over this period,

[00:30:43] whereas U.S. stocks had an expansion of a fee ratio.

[00:30:46] So they did enjoy some multiple free rating.

[00:30:49] Now, the way I interpret this is as follows,

[00:30:53] which is international stocks have underperformed U.S. stocks in terms of growth.

[00:30:57] And as a result, investors are extrapolating out the stagnation of international stocks

[00:31:02] into the future, right or wrong,

[00:31:04] which is why the multiples have not expanded the same way they have in the U.S.

[00:31:08] So I think it's kind of that.

[00:31:09] That's the causal mechanism, I believe, behind what's happened.

[00:31:12] But it's definitely been, you know, a double whammy, right?

[00:31:15] Where you get worse growth internationally, and then you get this kicker,

[00:31:18] which is the expectation of future worse growth.

[00:31:21] I guess the...

[00:31:23] Sorry.

[00:31:23] I was going to say, and I think to tie this back into the intangible thread, right,

[00:31:27] so we don't go off too off-kill thread here.

[00:31:29] So the big question, Andy, comes, if you believe what I just said,

[00:31:32] is that intangible...

[00:31:33] So is that growth is what matters, right?

[00:31:35] Where can we get growth within the EM space?

[00:31:37] And why is it that the EM, so that international stocks,

[00:31:40] have undergrown U.S. stocks?

[00:31:42] I did this work here, and it ties to the point I just made earlier, right,

[00:31:45] which is that it turns out that companies that invest in intangible assets

[00:31:49] have more growth in the future, right?

[00:31:51] This is the point I just made with regards to quality.

[00:31:54] And that's a pretty robust empirical finding.

[00:31:55] And it makes sense because intangible investment is the driver of growth today.

[00:32:01] That's just the fact.

[00:32:03] And it turns out that if you look across the world, now that I have my scores,

[00:32:09] that the non-U.S. countries have had, on average,

[00:32:15] much lower intangible investment than the U.S.

[00:32:17] So the U.S. score is the highest.

[00:32:18] There are some pockets of strength, say, in northern and central Europe,

[00:32:22] you know, Korea, Japan, Taiwan score are okay.

[00:32:25] But on average, the average international company is not investing as much today

[00:32:30] as the U.S. is in intangible assets.

[00:32:33] And more importantly, this pattern goes back through time, right?

[00:32:37] Because we are our scores today.

[00:32:38] We don't also backtest our scores all the way back to 95.

[00:32:41] And so one study I ran, which I thought was really cool in this paper, 677,

[00:32:46] is the scatterplot, where I show on the x-axis in 2010, the beginning of this sample period,

[00:32:54] on a country-by-country basis, what has been the average level of intangible value.

[00:32:58] And on the y-axis, I show over the subsequent 13 years,

[00:33:02] what has been the real EPS growth, earnings per share growth, in U.S. dollar terms.

[00:33:07] And I find, like, a 54-something percent correlation between those numbers, right?

[00:33:11] Which is really high.

[00:33:12] And then if you do a weighted regression, where you look at, like, the—

[00:33:15] you would give more weight to the U.S. than you do to, like, Malaysia,

[00:33:18] you get, like, a 70-something percent correlation.

[00:33:20] So the point here is that, you know, we can explain, like,

[00:33:23] the diversions in the wealth of nations over the past 13 years through one variable.

[00:33:27] Like, yes, it's the case that, like, there's so much cross-country variation,

[00:33:30] like, different legal systems, different cultures, different languages, so on and so forth.

[00:33:33] But at the end of the day, it all kind of boils down to the same thing,

[00:33:36] which is that these countries have, on average, invested less in R&D

[00:33:40] and in creating modern brands and, you know, building network effects.

[00:33:45] And that has led to comparative stagnation relative to the U.S.

[00:33:48] Why do you think that is?

[00:33:49] I mean, is it explained by the simple variable of we have way more technology companies,

[00:33:52] or is it a lot deeper than that?

[00:33:54] So we do have more technology companies, but I think the result is a lot deeper.

[00:33:58] If you look at the findings by sector, right, so you can look at, like, the 11 sectors.

[00:34:05] So yes, the U.S. has more technology companies, but within technology,

[00:34:08] our average technology company is more innovative or more intangible

[00:34:12] than the average one in Europe.

[00:34:14] And it's not just technology.

[00:34:16] It's finance.

[00:34:17] It's real estate.

[00:34:18] All 11 of the sectors, actually, the U.S. scores higher than international on intangibles.

[00:34:24] And so it's just the case that, you know, whatever sector you operate in,

[00:34:28] if you're in the U.S., you've been, on average, more willing to embrace kind of modern practices

[00:34:32] with regards to these things.

[00:34:34] So, you know, it's not fully set to reflect, but there's obviously an element to that too.

[00:34:38] So you think, what do you think the reason is then,

[00:34:39] that there's so much more investment in the U.S. in intangibles?

[00:34:42] I mean, that's a really good question.

[00:34:44] It's not really something that I deliberately tackled in this paper.

[00:34:47] My findings are more descriptive, like just saying it is what it is.

[00:34:50] This is like the way the world is.

[00:34:53] International stocks have underinvested in intangibles.

[00:34:56] But, you know, if I was just to kind of like venture a couple guesses,

[00:34:58] like I'd say that, you know, without getting to political, you know, capitalism,

[00:35:02] I think, you know, the U.S. is much more supportive of entrepreneurs

[00:35:05] and risk-taking and, you know, in that way, innovation, right?

[00:35:09] Like notwithstanding what Elon's doing with Doge, right?

[00:35:13] Like, you know, the U.S. on a comparative basis is still a lot more free

[00:35:17] and, you know, has fewer regulations than, say, Europe.

[00:35:19] Or you look at what they're doing with the AI,

[00:35:21] like they're basically not letting it happen,

[00:35:23] which is, you know, shooting yourself in the foot there.

[00:35:25] And I think, you know, another point that's kind of interesting too

[00:35:27] is like, you know, political systems is like, you know,

[00:35:30] Justin, you're going to appreciate this.

[00:35:32] Like I was talking with Perth Toll, our friend, about this a little while ago,

[00:35:36] you know, looking at the overlap between freedom scores

[00:35:38] and intangible value.

[00:35:40] And the correlation is pretty high.

[00:35:41] It's like 45% or something at the country level.

[00:35:45] And it kind of makes sense too, which is like, you know,

[00:35:47] especially on a human capital front, right?

[00:35:49] But if you want to encourage human capital formation,

[00:35:51] you know, literacy and the things that allow innovation and development,

[00:35:55] well, you need to have like an education system.

[00:35:57] You need to allow both economic and social freedoms.

[00:36:00] And so I think that's another component to this as well,

[00:36:01] that, you know, the U.S. is the most free country,

[00:36:04] both in terms of the capitalist side of things,

[00:36:07] but also just like in terms of, you know,

[00:36:08] the other kind of more intangible dimensions.

[00:36:11] One of the things we see a lot in the traditional financial statement world

[00:36:15] is that the U.S. data often is better than the data you can get internationally.

[00:36:20] Is that true here as well?

[00:36:21] Were you able to get as good data internationally,

[00:36:22] or did you have to make some compromises in terms of what you could do?

[00:36:25] Yeah, so that's definitely true.

[00:36:27] So, you know, one of the challenges with, you know,

[00:36:30] a traditional value strategy going international is accounting standards, right?

[00:36:34] Like as you go outside the U.S., like, you know,

[00:36:36] the accounting standards of different countries start to kind of be inconsistent, right?

[00:36:40] Which obviously creates problems,

[00:36:41] and that's, you know, one of the challenges that you deal with.

[00:36:44] One of the nice things about the intangible data sets that I look at

[00:36:48] is that they're kind of global, right?

[00:36:50] They're like, they're not constrained by border.

[00:36:52] It's not like each country has their own like LinkedIn or whatever, you know?

[00:36:56] And, you know, an example would be patents, you know, we shouldn't go back to,

[00:36:59] which is if you look at the top 10 patent holders of U.S. patents, in this case, right?

[00:37:04] Only four of them are U.S. companies, right?

[00:37:07] Six out of the 10 top patent holders, including number one by far, which is Samsung,

[00:37:12] are not U.S.

[00:37:13] And the point being that, like, we all operate in a global market these days,

[00:37:17] regardless of your domicile.

[00:37:18] And so a lot of these data sets are, you know, quite consistent and universal.

[00:37:22] Now, granted, there are coverage differences, especially as you go out to EM,

[00:37:27] where, you know, fewer people might be using LinkedIn in Malaysia, right,

[00:37:33] than, like, India.

[00:37:33] And there's a lot of kind of random cross-section, cross-ventory variation.

[00:37:36] But that's, at least from an alpha standpoint, offset by the fact that,

[00:37:40] you know, there's also just more inefficiency in these international markets in general.

[00:37:44] And so you don't need as good hovers.

[00:37:45] You don't need as much signal noise in order to generate returns.

[00:37:50] So, you know, that's just my findings so far.

[00:37:53] That's interesting.

[00:37:54] Samsung has the most patents.

[00:37:55] I mean, Justin, I probably would have guessed Apple, maybe,

[00:37:56] if I had to guess who had the most patents.

[00:37:57] I'm always amazed at these patent counts because these companies and, like,

[00:38:01] IBM's always up there, too, and stuff.

[00:38:03] Yeah.

[00:38:04] Yeah.

[00:38:04] Yeah.

[00:38:05] IBM was number one for, like, famously for, like, over a decade

[00:38:08] until they were superseded by Samsung.

[00:38:10] And now it looks like it's kind of the beginning of the end.

[00:38:12] It's kind of a symbolic turn, right, where, like,

[00:38:15] you look at the patent share of U.S. patents,

[00:38:16] it's, like, used to be, you know, almost all U.S.

[00:38:19] and then the rest of the world.

[00:38:20] And then what's happened is that, like, five Asian countries,

[00:38:22] China, Taiwan, Korea, and maybe India,

[00:38:26] have cut into the market share of the U.S.,

[00:38:28] taking a big share.

[00:38:29] The rest of the world, Europe, et cetera, has been flat.

[00:38:31] And the U.S. has kind of ceded to the Chinas and Taiwans of the world.

[00:38:36] And I think, you know, you don't want to read too much into this.

[00:38:38] This is just representative of one type of IP,

[00:38:41] which is just one of the four pillars.

[00:38:43] But I think it is symbolic of the fact that, you know,

[00:38:46] the U.S. isn't the only game in town anymore.

[00:38:48] You know, in this global world, yes, you know,

[00:38:51] we are the leader if you only can pick one country.

[00:38:53] But if you could pick more than one country,

[00:38:55] you know, you're starting to see inroads

[00:38:57] in terms of the evolution of other countries

[00:39:00] with regards to investments in this area.

[00:39:03] Yeah, and I would guess, like,

[00:39:05] firms that make physical devices probably need more patents.

[00:39:07] So just thinking about Samsung having the most,

[00:39:08] that's probably part of this too.

[00:39:10] Yes, it's easier to patent that than, like, you know,

[00:39:12] more kind of, you know, software type stuff, yeah.

[00:39:16] So going off our tangent, which we always get on this podcast,

[00:39:19] and going back to intangible value,

[00:39:21] just one more from you before I hand it back to Justin.

[00:39:22] You looked at, you actually have a great chart,

[00:39:25] which we'll put in the podcast,

[00:39:26] about this underperformance of international stocks.

[00:39:28] And you looked at a factor where you just invested

[00:39:31] in the highest intangible companies

[00:39:33] and what that did relative to the underperformance

[00:39:35] of international stocks.

[00:39:36] So can you talk about that?

[00:39:37] Yeah, so I think I mentioned that, you know,

[00:39:39] the setup of this paper was showing that,

[00:39:42] on a, you know, real overinflation basis,

[00:39:45] international index had only delivered

[00:39:47] two and a half percentage points per year

[00:39:49] versus, like, 10% for the US.

[00:39:50] So there was a seven and a half ring gap.

[00:39:52] And so it turns out that if you don't just buy

[00:39:55] the median international stock,

[00:39:58] but take advantage of the fact

[00:39:59] that there's massive dispersion, right,

[00:40:01] and just buy the subset of international stocks

[00:40:02] that are high intangibles,

[00:40:04] the Samsungs of the world, the, you know,

[00:40:07] the LVMHs, the Novo Nordists.

[00:40:10] And it turns out that the portfolio you build

[00:40:12] would have actually outperformed the index significantly

[00:40:15] by, I think, about four and a half percentage points

[00:40:17] per year, taking you to almost eight percentage points

[00:40:19] per year.

[00:40:20] And then if you adjust for the currency effect, right,

[00:40:22] which is that the US dollar is appreciated

[00:40:24] on a relative basis over this period,

[00:40:26] you end up with returns that are basically the same

[00:40:29] as what the US index has delivered since 2010.

[00:40:32] And I think this is kind of an interesting finding

[00:40:35] because, you know, everyone talks about how, you know,

[00:40:38] Europe and international stocks have underperformed

[00:40:41] because they're just worse.

[00:40:42] International stocks are bad.

[00:40:44] US stocks are good.

[00:40:45] But like, it's not about US versus international.

[00:40:47] It's about intangible versus tangible.

[00:40:49] What's happened is that the intangible economy

[00:40:50] has taken off and the tangible economy has stagnated.

[00:40:53] And, you know, international,

[00:40:55] the international countries have, for whatever reason,

[00:40:58] we've discussed a few of them,

[00:40:59] decided to, you know, have this focus

[00:41:02] on the kind of more legacy industries.

[00:41:04] And as a result, they've kind of hitched their wagon

[00:41:06] to the wrong horse,

[00:41:07] missed the growth of the intangible economy.

[00:41:09] But it's merely an artifact of that mix,

[00:41:11] less specific to the countries themselves, right?

[00:41:15] Because as we just showed,

[00:41:18] the countries, the companies within the international index

[00:41:20] that are highly intangible have done just fine.

[00:41:22] Funny, like this is the variable

[00:41:24] that explains everything, by the way.

[00:41:25] Like that's the way you almost look at it.

[00:41:27] Like values underperformed, adjust for intangibles,

[00:41:29] it doesn't underperform anymore.

[00:41:30] Like international is underperformed,

[00:41:31] adjust for intangibles,

[00:41:32] it doesn't underperform anymore.

[00:41:33] So, I mean, this is a huge part of the answer

[00:41:34] to a lot of these questions we've been asking, I think.

[00:41:36] And small caps as well in the U.S.

[00:41:40] So in that chart, you have the excess return.

[00:41:45] So my only request is,

[00:41:46] can we have excess return podcast?

[00:41:48] And I can link to our YouTube channel over there,

[00:41:52] give us a little cross promotion.

[00:41:54] Back to the patents, Justin,

[00:41:55] we should be having excess returns,

[00:41:56] which clearly never going to happen,

[00:41:58] but that way, whenever they use it,

[00:41:59] they have to use us.

[00:42:00] Well, maybe we were somehow in the back of Kai's mind

[00:42:02] when he can put that excess return line in there.

[00:42:05] I don't know.

[00:42:09] And what is the contributing factor to that outperformance?

[00:42:13] Is it stock selection?

[00:42:14] Is it industry variation?

[00:42:17] I mean, what did you get?

[00:42:19] Yeah.

[00:42:20] So I think we talked a little bit about industries earlier

[00:42:23] and kind of my answer was,

[00:42:25] it's part but not the full explanation.

[00:42:26] I think the answer is here,

[00:42:27] the same as well.

[00:42:28] So what you can do is you can decompose the returns,

[00:42:31] the excess returns of the portfolio

[00:42:34] in two different components.

[00:42:35] That derived from sector selection, right?

[00:42:39] And that derived from stock selection.

[00:42:42] In other words, the residual, right?

[00:42:44] If you look at the sector composition of the portfolio,

[00:42:47] it tends to be, as you might expect,

[00:42:50] undoing a lot of the biases of the international index

[00:42:52] as opposed to the US.

[00:42:53] So whereas the international stocks tend to be

[00:42:55] a lot of banks and industrials

[00:42:57] and a few tech companies,

[00:42:59] the intangible portfolio tends to be more tech,

[00:43:02] more healthcare, a lot less banks.

[00:43:04] And so that bias, you know, obviously with hindsight,

[00:43:06] we know was a contributor to returns.

[00:43:08] But it turns out it's not that big of a contributor.

[00:43:11] Ends up being about one third of the returns,

[00:43:14] whereas two thirds still come from stock selection.

[00:43:16] In other words, picking the best banks

[00:43:17] within the banking sector

[00:43:18] or picking the best technology companies

[00:43:20] within the technology sector.

[00:43:22] And I think the other point to mention in this

[00:43:23] is that, you know, obviously the word hindsight

[00:43:26] is kind of the operative word.

[00:43:27] It's doing a lot of heavy lifting here,

[00:43:28] which is if you said back in like 2007,

[00:43:30] right, and you're saying,

[00:43:31] hey, what sector should I buy?

[00:43:32] I mean, financials had just done really well.

[00:43:34] And so you're sitting there,

[00:43:35] you know, maybe you don't really know.

[00:43:37] And, you know, obviously you could try

[00:43:38] to form some kind of projection

[00:43:39] based on macro forecasting.

[00:43:41] But, you know, I think another way of doing it,

[00:43:43] and it might have been a better way of doing it,

[00:43:45] is just to go organically bottoms up

[00:43:46] and say, which companies are acquiring the best talent,

[00:43:50] right, getting the most, you know, trademarks, et cetera.

[00:43:54] And that's where the innovation is

[00:43:55] and just buy those companies.

[00:43:56] And from that emerges bottoms up again,

[00:43:59] like a sector allocation.

[00:44:00] And it just so happens that

[00:44:01] they might be more tech companies,

[00:44:03] might be more former companies

[00:44:05] and fewer banks.

[00:44:06] But that was the right answer.

[00:44:07] So you got to the right answer

[00:44:08] without having to kind of force anything top down.

[00:44:12] What about the country composition?

[00:44:15] I mean, you mentioned that you talked to Perth

[00:44:16] in sort of freer countries,

[00:44:17] and we'll talk about emerging markets in a minute,

[00:44:19] but what did you find

[00:44:20] with intangible value international

[00:44:22] versus like your standard international benchmark?

[00:44:25] Yeah, so I think that the overweights

[00:44:26] on a country level were to Germany,

[00:44:29] France, Japan, I think Taiwan.

[00:44:31] And the underweights were to like China,

[00:44:35] Australia, Canada, Saudi, Brazil.

[00:44:38] And that kind of maps to

[00:44:39] what you would kind of intuitively understand

[00:44:42] is the point of intangible value.

[00:44:43] So on the underweight side, it's pretty obvious.

[00:44:45] You know, Australia and Canada

[00:44:46] are big commodity exporters.

[00:44:48] That's more acid-heavy, tangible businesses

[00:44:51] that, you know, are not, you know,

[00:44:52] considered attractive here.

[00:44:53] On the overweight side, you know,

[00:44:55] Germany is, despite its problems,

[00:44:58] you know, a leader in kind of industrial manufacturing.

[00:45:03] I'd say France, you know,

[00:45:04] a lot of the companies are more

[00:45:05] on kind of the brand equity side.

[00:45:07] You know, Japan, Taiwan, South Korea,

[00:45:10] leaders in kind of semi-dub conductors

[00:45:11] and precision manufacturing as well,

[00:45:13] which require a little IP.

[00:45:15] So, yeah.

[00:45:17] So I think that kind of maps to intuition here.

[00:45:20] And how would you...

[00:45:22] So that was on developed international,

[00:45:24] and then you basically applied the same...

[00:45:26] We could take the same discussion

[00:45:28] and like have it over emerging markets.

[00:45:31] What were the takeaways there?

[00:45:34] Yeah.

[00:45:34] So I think, first of all,

[00:45:35] the reason why I wanted to run this analysis separate

[00:45:37] was because, you know, by definition,

[00:45:39] emerging markets are, you know,

[00:45:42] less developed than developed markets

[00:45:43] and, you know, therefore more kind of

[00:45:45] in the old economy

[00:45:46] and are thus more reliant on tangible equity.

[00:45:49] Like what, to take a side tangent,

[00:45:51] in the US, one thing I have found years ago

[00:45:54] is that if you run traditional value factors

[00:45:56] in only the most tangible industries,

[00:45:58] they do okay.

[00:45:59] It's only when you start trying to apply those factors

[00:46:02] into the intangible side of the sectors,

[00:46:05] let's call it,

[00:46:06] you start to underperform.

[00:46:07] Right?

[00:46:07] Like trying to use price of book

[00:46:09] to predict Google versus Amazon

[00:46:10] is not going to be helpful,

[00:46:11] but trying to do it between two banks

[00:46:12] might actually work,

[00:46:13] which is what I find.

[00:46:14] Now, of course, the problem in the US

[00:46:15] is that the intangible sectors

[00:46:17] are now 80% of the market

[00:46:18] and increasing, right?

[00:46:20] So, you know, at some point,

[00:46:21] there's nothing to pick from.

[00:46:22] But going back to the point here is,

[00:46:24] you know, that was the concern with EM,

[00:46:26] which is maybe you don't need any of this.

[00:46:27] Maybe you just do price of the book

[00:46:28] and you're done, right?

[00:46:30] Or price of earnings or price of sales, whatever.

[00:46:31] You don't need intangibles.

[00:46:34] So, you know, I decided I'd run the analysis,

[00:46:36] did the exact same portfolio construction

[00:46:38] as we did in developed in US,

[00:46:40] which is, you know, take the universe of stocks,

[00:46:42] select the top, you know, 10,

[00:46:44] I think it was top 150 stocks,

[00:46:45] and then look at that compared to the market.

[00:46:49] And in this case,

[00:46:50] the performance was actually a little better

[00:46:51] on a relative basis,

[00:46:53] mainly because the benchmark, right?

[00:46:55] The index was worse.

[00:46:56] The EM index has done really poorly,

[00:46:58] right?

[00:46:58] The past 10 or so years.

[00:47:01] You know, one of the things I looked at

[00:47:02] that was quite interesting was,

[00:47:04] so if you're an EM fund manager,

[00:47:06] basically only one thing has mattered

[00:47:07] for the past few years,

[00:47:08] which is like,

[00:47:09] what has been your view on China, right?

[00:47:11] That's been the entire game.

[00:47:12] You know, China was at 1.40% of the MSI index.

[00:47:15] It then went on to underperform at 43%.

[00:47:18] And, you know,

[00:47:20] those managers who were underweight China

[00:47:22] look like heroes,

[00:47:23] and those who are not look like idiots.

[00:47:25] And now we're seeing the launch

[00:47:26] of like a million EMF China funds, right?

[00:47:29] And so the question I wanted to ask was,

[00:47:32] you know,

[00:47:32] can intangible value get you to the same place,

[00:47:34] but organically, right?

[00:47:35] Without kind of making any kind of macro views.

[00:47:38] And it was really interesting

[00:47:39] because it turns out that the exposure

[00:47:41] of the intangible value portfolio to China

[00:47:44] ranged through time

[00:47:45] from I think negative 20% to 0%.

[00:47:47] So it was on average underweight,

[00:47:49] but like changed.

[00:47:50] And that generally mirrored the inverse

[00:47:52] of the price

[00:47:53] or the relative returns of China

[00:47:54] because this is a value strategy.

[00:47:56] So when China rallies,

[00:48:00] you hold less.

[00:48:01] And when China falls on a relative basis,

[00:48:04] you hold more.

[00:48:05] And interestingly,

[00:48:06] when China was at its peak in 2020,

[00:48:09] this strategy was at its max underweight.

[00:48:10] It was like 20 something percent

[00:48:12] underweight China.

[00:48:13] And then as China then collapsed,

[00:48:15] which it did,

[00:48:17] you know,

[00:48:17] it kind of like took profits

[00:48:18] and is now kind of more at a neutral weight.

[00:48:20] So that was really interesting

[00:48:22] that you're able to kind of get

[00:48:23] to the same point

[00:48:23] simply by looking at intangible assets.

[00:48:26] And then the final thing

[00:48:27] I wanted to look at,

[00:48:27] obviously that's related,

[00:48:28] is to be sure

[00:48:29] that this is not just a fluke.

[00:48:31] In other words,

[00:48:31] you didn't just make one lucky bet

[00:48:33] and that's why you have excess returns.

[00:48:35] Maybe it's the case.

[00:48:36] You know,

[00:48:36] how can you test that?

[00:48:38] So I did the same analysis

[00:48:39] as before with the sectors,

[00:48:40] but it's on the countries

[00:48:41] and said,

[00:48:42] let's attribute the returns

[00:48:44] of this strategy

[00:48:45] and excess of the benchmark

[00:48:46] to that derived,

[00:48:48] that which is derived

[00:48:49] from overweights

[00:48:51] to various countries

[00:48:52] and then the residual.

[00:48:53] So stock picking alpha.

[00:48:54] And it turns out

[00:48:55] that similar,

[00:48:56] the split was almost

[00:48:57] the exact same as before,

[00:48:58] that, you know,

[00:48:59] two thirds plus

[00:48:59] was coming from stock selection

[00:49:01] and only one third

[00:49:02] was coming from country selection.

[00:49:04] And then even if you remove

[00:49:05] the China thing

[00:49:05] and just manually get rid of it,

[00:49:07] you're still seeing,

[00:49:07] you know,

[00:49:08] some outperformance

[00:49:08] from country selection.

[00:49:10] So that was a pretty

[00:49:11] encouraging result.

[00:49:13] And, you know,

[00:49:13] kind of the point here

[00:49:15] with the emerging

[00:49:15] was just to say,

[00:49:16] like, let's just like try to,

[00:49:18] you know,

[00:49:18] get the most external

[00:49:19] validity as we can

[00:49:19] from this intangible value concept.

[00:49:21] Again, we started in the US

[00:49:22] and now we're bringing it

[00:49:23] to developed international

[00:49:24] and now emerging.

[00:49:25] So we have like three different

[00:49:27] kind of portfolios.

[00:49:28] And it's really interesting

[00:49:29] that, you know,

[00:49:30] first of all,

[00:49:30] it works in all three areas

[00:49:31] on a relative basis.

[00:49:32] That's nice.

[00:49:34] But what's also interesting

[00:49:35] is that the correlations

[00:49:36] between these three portfolios

[00:49:38] are pretty low,

[00:49:39] like somewhere between 25 and 50%.

[00:49:41] So lower than you might expect

[00:49:42] given that they're doing

[00:49:43] the identical thing,

[00:49:44] right,

[00:49:44] from a methodological standpoint.

[00:49:46] And, you know,

[00:49:47] it makes sense

[00:49:47] because there's zero overlap

[00:49:48] by definition

[00:49:49] in the holdings, right?

[00:49:51] Like the EM portfolio

[00:49:52] tends to hold,

[00:49:53] you know,

[00:49:53] a bit more tech.

[00:49:54] US holds a bit more tech,

[00:49:55] developed towards more brand.

[00:49:57] And so they end up

[00:49:58] having different compositions.

[00:49:59] So I think there's some value

[00:50:00] there in kind of

[00:50:01] the diversification

[00:50:02] of, you know,

[00:50:03] appliance factor

[00:50:04] across, you know,

[00:50:05] these different regions.

[00:50:06] You know,

[00:50:07] and as we mentioned,

[00:50:07] this isn't just like

[00:50:08] a research paper now.

[00:50:09] There's actually a physical,

[00:50:10] you know,

[00:50:11] investable product

[00:50:12] via an ETS

[00:50:13] that is launched

[00:50:14] that is investing

[00:50:15] in these international firms

[00:50:18] that you're identifying.

[00:50:19] So hopefully it's a great time

[00:50:23] to be launching

[00:50:23] an international fund.

[00:50:26] Yeah,

[00:50:26] I mean,

[00:50:27] look,

[00:50:27] I think it is.

[00:50:28] I mean,

[00:50:28] we all know

[00:50:29] that the bull case

[00:50:30] for international stocks

[00:50:32] that,

[00:50:32] you know,

[00:50:32] US has outperformed.

[00:50:33] It's, you know,

[00:50:34] due to the,

[00:50:35] you know,

[00:50:35] an increasingly concentrated index.

[00:50:37] You're at 35% now

[00:50:38] of the index in the S&P

[00:50:39] is behind seven stocks.

[00:50:40] These stocks are more

[00:50:41] and more inflated

[00:50:43] in their valuations.

[00:50:44] And therefore,

[00:50:44] we're looking for ways

[00:50:45] to diversify,

[00:50:46] you know,

[00:50:46] either within the US

[00:50:47] but even better

[00:50:48] just outside the US.

[00:50:50] And, you know,

[00:50:51] non-US stocks

[00:50:51] traded a 50% discount

[00:50:52] on Shiller PE

[00:50:53] to US stocks.

[00:50:55] Obviously,

[00:50:55] the average international stock,

[00:50:57] as I've said,

[00:50:58] is less,

[00:50:59] you know,

[00:51:00] focused on innovation

[00:51:00] and intangible investment.

[00:51:01] And thus,

[00:51:02] we expect to grow less fast.

[00:51:04] But again,

[00:51:04] you just subset the stocks

[00:51:05] that are,

[00:51:06] you know,

[00:51:06] the, you know,

[00:51:07] still high in intangibles

[00:51:08] and you end up

[00:51:09] with a portfolio

[00:51:10] that, you know,

[00:51:10] is in theory,

[00:51:11] we hope,

[00:51:12] you know,

[00:51:12] has similar intangible exposure

[00:51:14] as the US stocks

[00:51:16] yet trades at this kind of

[00:51:17] baby with a bathwater

[00:51:18] type thing

[00:51:19] at this, you know,

[00:51:20] pretty significant discount

[00:51:22] on kind of

[00:51:22] traditional PE metrics.

[00:51:23] So, you know,

[00:51:24] if all goes well,

[00:51:25] you're getting the best

[00:51:25] of both worlds scenario.

[00:51:28] What do you think

[00:51:28] of the idea of using,

[00:51:30] and we know a few guys

[00:51:31] that are doing this,

[00:51:32] like using these LMs

[00:51:33] to actually,

[00:51:34] you know,

[00:51:35] you give it the prompt

[00:51:36] about the type of investor

[00:51:37] that it's seeking to

[00:51:39] model or replicate.

[00:51:40] You know,

[00:51:41] it's basically

[00:51:43] scouring the internet

[00:51:44] and all the financial data

[00:51:46] to then build a portfolio

[00:51:47] that has the characteristics

[00:51:51] that the system

[00:51:52] has been given

[00:51:53] about what types of stocks

[00:51:54] to look for

[00:51:55] that could contribute

[00:51:56] to market outperformance.

[00:51:57] Do you have any thoughts

[00:52:00] on like using

[00:52:02] the technology

[00:52:02] in that way

[00:52:03] or is that like,

[00:52:05] is that going,

[00:52:06] is that like too much

[00:52:07] of just one big

[00:52:08] backtest

[00:52:09] and you're kind of

[00:52:09] just data mining

[00:52:10] the data mining?

[00:52:12] I don't know.

[00:52:13] What are your thoughts?

[00:52:15] Yeah, so I think

[00:52:16] there's two questions

[00:52:16] in there.

[00:52:17] And the first one is

[00:52:17] would you consider

[00:52:18] like a strategy

[00:52:20] about uses

[00:52:20] large language models

[00:52:21] or AI

[00:52:21] to look for stocks

[00:52:23] that have like,

[00:52:24] so looking for things

[00:52:25] that have historically

[00:52:26] done well,

[00:52:26] have backtest it well.

[00:52:27] So say,

[00:52:28] you know,

[00:52:28] I want to find me

[00:52:29] all the things

[00:52:29] that like have historically

[00:52:31] had higher returns

[00:52:32] in the future, right?

[00:52:33] I'm pretty skeptical

[00:52:34] of that approach.

[00:52:35] I think there's a few

[00:52:36] problems with it,

[00:52:38] namely around like,

[00:52:39] you know,

[00:52:39] as quants,

[00:52:40] we like to think

[00:52:40] that we live

[00:52:41] in a big data environment.

[00:52:42] We don't, right?

[00:52:43] You think about it this way,

[00:52:44] which is,

[00:52:45] you know,

[00:52:45] there are 3,000 US stocks,

[00:52:46] 30 years of data,

[00:52:47] so that's 90,000 data points.

[00:52:50] LAMA was trained

[00:52:51] on 3 trillion tokens, right?

[00:52:53] So like,

[00:52:53] that's so much less data.

[00:52:55] And not only that,

[00:52:56] but the data is more noisy, right?

[00:52:58] So like,

[00:52:59] you know,

[00:52:59] if you get a 51% hit rate,

[00:53:00] you're a pretty good investor.

[00:53:02] So that effectively dilutes

[00:53:03] the informational content

[00:53:04] of each data point.

[00:53:05] And then the final thing,

[00:53:06] I think the most important thing

[00:53:07] is the non-stationarity

[00:53:09] that, you know,

[00:53:10] markets are evolving

[00:53:11] through time.

[00:53:11] Things change,

[00:53:12] reenter new regimes

[00:53:14] and anomalies

[00:53:15] that may have worked

[00:53:15] in the past,

[00:53:16] they get arbitraried away.

[00:53:18] And so anything that attempts

[00:53:19] to fit on historical data patterns

[00:53:20] to say,

[00:53:21] use this,

[00:53:23] use this,

[00:53:23] this like really fancy model

[00:53:25] to try to predict,

[00:53:27] you know,

[00:53:27] which stocks

[00:53:28] have been correlated

[00:53:28] with the highest returns

[00:53:29] and then use that

[00:53:30] to predict the future.

[00:53:30] Those things tend to,

[00:53:32] I think,

[00:53:32] be a little bit overfitted

[00:53:33] or a lot overfitted.

[00:53:34] I'd be concerned

[00:53:35] about an approach that way.

[00:53:36] That being said,

[00:53:37] the other part of your question

[00:53:38] was pretty interesting,

[00:53:39] which was like,

[00:53:40] if you want to say,

[00:53:41] so you can ask

[00:53:42] Shaki Bikini and say,

[00:53:42] hey,

[00:53:43] here are all my,

[00:53:44] here are a bunch of papers

[00:53:44] I've written.

[00:53:45] Now write in the style

[00:53:46] of Kai Wu,

[00:53:47] this thing,

[00:53:47] write a paper in the style

[00:53:49] of Kai,

[00:53:49] or,

[00:53:49] you know,

[00:53:50] like similar to what

[00:53:51] Majority does

[00:53:51] or the new video models,

[00:53:54] right?

[00:53:54] Create a animation

[00:53:55] or a image

[00:53:56] in a style of Van Gogh,

[00:53:58] right?

[00:53:58] You can do that.

[00:53:59] It's like,

[00:53:59] you know,

[00:53:59] call style transfer.

[00:54:00] So to do style transfer

[00:54:02] on a investment standpoint,

[00:54:03] might make sense,

[00:54:04] right?

[00:54:04] Because you're not now

[00:54:05] trying to overfit

[00:54:06] on historical market data,

[00:54:08] which is not meaningful,

[00:54:09] but you're instead

[00:54:10] trying to say,

[00:54:10] let's extrapolate

[00:54:11] from Warren Buffett's,

[00:54:12] you know,

[00:54:13] a hundred or so letters

[00:54:13] or whatever he has,

[00:54:15] a style

[00:54:15] and try to use that

[00:54:16] to move forward.

[00:54:17] So I actually started

[00:54:18] running a paper on this

[00:54:19] called like the Super Investors

[00:54:20] and like they had

[00:54:20] like a cover

[00:54:21] that would be Avengers,

[00:54:22] right?

[00:54:22] Take all the best investors

[00:54:23] and like,

[00:54:24] you know,

[00:54:24] encapsulate them

[00:54:25] in an L&M.

[00:54:26] It didn't really go anywhere

[00:54:27] because,

[00:54:28] you know,

[00:54:28] the limitation

[00:54:29] is context windows,

[00:54:30] right?

[00:54:30] So like,

[00:54:31] you know,

[00:54:32] at least when I was

[00:54:32] writing this paper,

[00:54:33] which is a little while ago,

[00:54:34] you couldn't fit all

[00:54:35] Buffett's letters

[00:54:36] into a single

[00:54:36] like context window.

[00:54:38] And so you would need

[00:54:38] to fine tune

[00:54:39] on this sort of data,

[00:54:40] which would be,

[00:54:41] you know,

[00:54:41] expensive and time-consuming.

[00:54:42] So I kind of didn't really

[00:54:43] go anywhere with it.

[00:54:44] But,

[00:54:45] you know,

[00:54:45] I think that there's

[00:54:46] at least from a kind of

[00:54:47] like theoretical standpoint,

[00:54:48] I think that avenue

[00:54:49] has more promise,

[00:54:51] right?

[00:54:51] Like,

[00:54:51] you know,

[00:54:52] one common thing

[00:54:52] in like the pod shop world,

[00:54:53] right,

[00:54:54] is this idea of like

[00:54:55] alpha capture

[00:54:55] where you take like

[00:54:56] your best PMs,

[00:54:58] like trades

[00:54:58] and their investment memos

[00:54:59] and you basically try

[00:55:00] to like replicate them

[00:55:01] and scale them up

[00:55:02] in the center book,

[00:55:02] right?

[00:55:03] So this is kind of

[00:55:03] the same idea as that,

[00:55:05] but saying,

[00:55:05] all right,

[00:55:05] like let's take,

[00:55:07] let's do this,

[00:55:08] but we'll use LLMs this time

[00:55:09] to kind of,

[00:55:10] you know,

[00:55:11] capture the alpha

[00:55:11] of these managers.

[00:55:12] So again,

[00:55:13] it presupposes

[00:55:14] being able to identify

[00:55:15] exactly who the managers are,

[00:55:16] which is,

[00:55:17] you know,

[00:55:17] itself a challenge,

[00:55:18] but in theory,

[00:55:19] I think it could be

[00:55:19] a viable approach.

[00:55:21] If you resurrect that,

[00:55:23] if you resurrect

[00:55:24] that super investors paper,

[00:55:25] can we get first dibs

[00:55:26] on you coming on the podcast?

[00:55:28] Okay.

[00:55:29] Yeah.

[00:55:30] Nice.

[00:55:32] I've also wondered,

[00:55:33] are they biased

[00:55:34] like towards larger companies?

[00:55:35] So for instance,

[00:55:36] you would never actually do

[00:55:36] this in the real world,

[00:55:37] but if you ever say

[00:55:38] to chat GPT,

[00:55:39] you know,

[00:55:39] give me the 10 best stocks

[00:55:40] to buy now,

[00:55:41] it always seems like

[00:55:42] they have more information.

[00:55:43] So it always gives you

[00:55:44] like a bunch of large cap stocks back.

[00:55:45] And I don't know,

[00:55:46] I don't know enough

[00:55:47] about how they work

[00:55:47] to say if that's true,

[00:55:48] but I do wonder

[00:55:49] if like it would ever

[00:55:50] give you any small cap stocks

[00:55:51] or if it's always going

[00:55:52] to give you the companies

[00:55:52] that has the most information on.

[00:55:54] You could probably ask it

[00:55:55] to give you only small cap stocks

[00:55:56] or if you've trained it

[00:55:57] only on like the whole,

[00:55:58] only on the letters

[00:55:59] of like a micro cap investor,

[00:56:01] you know,

[00:56:01] maybe you'd only get

[00:56:01] small cap stocks.

[00:56:03] But yeah,

[00:56:03] by default,

[00:56:04] it's not going to do that.

[00:56:06] Right.

[00:56:06] Because if,

[00:56:07] you know,

[00:56:07] all this LLM they're doing

[00:56:08] is they're just like

[00:56:09] next word prediction.

[00:56:09] They're just looking at like

[00:56:10] frequencies and probabilities.

[00:56:11] And it turns out to be the case

[00:56:13] that a lot more people

[00:56:14] talk about Tesla

[00:56:14] and NVIDIA

[00:56:15] than they do about,

[00:56:16] you know,

[00:56:17] some small cap.

[00:56:18] And as a result,

[00:56:19] they're trying to give you

[00:56:19] what you want.

[00:56:20] Because if you don't tell it

[00:56:21] specifically what you want,

[00:56:22] otherwise,

[00:56:22] it's going to assume

[00:56:23] what you want

[00:56:23] is to hear more about NVIDIA.

[00:56:25] All right.

[00:56:26] Standard closing question

[00:56:27] for you.

[00:56:28] Sorry,

[00:56:28] it's such a hard pivot,

[00:56:29] but we'd like to ask our guests

[00:56:32] one standard closing question.

[00:56:33] This is a new one for you.

[00:56:34] So what is something

[00:56:34] you believe

[00:56:35] about investing

[00:56:36] that most of your peers

[00:56:37] would disagree with you on?

[00:56:40] Yeah,

[00:56:40] so we can kind of

[00:56:41] continue on this AI theme

[00:56:42] because why not?

[00:56:44] So,

[00:56:47] you know,

[00:56:47] a lot of folks

[00:56:48] are pretty like

[00:56:50] concerned about

[00:56:50] AI taking our jobs,

[00:56:52] not just in finance,

[00:56:53] but, you know,

[00:56:53] at least in our industry,

[00:56:54] you know,

[00:56:54] portfolio managers,

[00:56:55] financial analysts.

[00:56:56] So I actually wrote

[00:56:57] a paper on this.

[00:56:58] I actually wrote a few papers

[00:56:59] on this topic,

[00:57:00] you know,

[00:57:01] of how will AI LLMs

[00:57:03] impact our industry

[00:57:04] and our jobs,

[00:57:05] right?

[00:57:06] And I think,

[00:57:06] you know,

[00:57:06] the first thing to note

[00:57:07] is what I just told you guys,

[00:57:08] you know,

[00:57:09] just now,

[00:57:10] which is that,

[00:57:10] you know,

[00:57:11] I think that

[00:57:11] trying to use

[00:57:13] large language models

[00:57:13] to replace the kind of

[00:57:14] capital allocation component

[00:57:16] of investing.

[00:57:17] So in other words,

[00:57:18] hey,

[00:57:18] here's factors you

[00:57:19] have a thousand factors,

[00:57:20] find the best ones

[00:57:21] based on historical

[00:57:23] correlation with returns

[00:57:24] is non-starter.

[00:57:25] I think that's actually

[00:57:25] not the way to do it.

[00:57:27] And I've written,

[00:57:28] you know,

[00:57:28] papers on this,

[00:57:29] on why not.

[00:57:32] I think that the killer

[00:57:33] use case of AI,

[00:57:35] and this was the 2020 paper

[00:57:36] I wrote on,

[00:57:37] where I said,

[00:57:38] you don't want to do

[00:57:39] the first thing,

[00:57:40] but what you do want to do

[00:57:41] is to use large language models

[00:57:43] as a way of structuring

[00:57:44] unstructured data.

[00:57:45] You know,

[00:57:45] I specifically called out

[00:57:46] this technology

[00:57:47] and said that this is

[00:57:48] the killer use case.

[00:57:49] And I think,

[00:57:49] you know,

[00:57:50] over the past five or so years

[00:57:51] since that paper came out,

[00:57:52] it's basically become

[00:57:53] like common knowledge,

[00:57:54] right?

[00:57:54] Most people will agree

[00:57:55] that what large language models

[00:57:56] are doing is,

[00:57:58] you know,

[00:57:58] working with unstructured data.

[00:58:00] So now to take this

[00:58:01] to my point is,

[00:58:03] you know,

[00:58:03] in my last paper

[00:58:04] on AI financial analysts,

[00:58:06] one thing I did

[00:58:07] was I said,

[00:58:07] let's look at the job

[00:58:08] of a financial analyst

[00:58:09] or a PM.

[00:58:10] And you can decompose it

[00:58:12] into, say,

[00:58:12] 20 or 30 different tasks.

[00:58:14] These are individual things

[00:58:15] like, you know,

[00:58:16] creating PowerPoints

[00:58:17] or mapping Q6

[00:58:18] or whatever, right?

[00:58:21] And it turns out

[00:58:22] that you can then ask

[00:58:24] the LLM

[00:58:24] or figure out

[00:58:25] which of these

[00:58:25] individual tasks

[00:58:26] are most

[00:58:27] or better accomplished

[00:58:28] using a large language model

[00:58:29] and which are better

[00:58:30] accomplished

[00:58:30] using a human.

[00:58:31] And the results

[00:58:32] are intuitive, right?

[00:58:33] Talking to clients,

[00:58:33] building your business,

[00:58:34] that's human.

[00:58:35] Using creativity

[00:58:36] with new investment strategies,

[00:58:38] that's human.

[00:58:41] Creating PowerPoints,

[00:58:42] that's better

[00:58:43] by machines, right?

[00:58:44] Proofing, right?

[00:58:45] And so it turns out

[00:58:46] that about half

[00:58:47] of the tasks

[00:58:47] of an analyst today

[00:58:49] are better

[00:58:50] with large language models

[00:58:51] and half are with humans.

[00:58:53] And so you think

[00:58:54] about what our jobs are.

[00:58:55] They're just bundles

[00:58:55] of tasks.

[00:58:56] I don't see us

[00:58:57] kind of like

[00:58:58] net messages

[00:58:58] using jobs.

[00:58:59] I just see like

[00:59:00] a repackaging

[00:59:00] of those jobs, right?

[00:59:01] You almost think

[00:59:02] of like a new AI

[00:59:03] enters the workforce.

[00:59:05] They take up

[00:59:05] the things that

[00:59:06] they're better at

[00:59:06] and you're left

[00:59:07] with the things

[00:59:07] that you're better at.

[00:59:09] Which, you know,

[00:59:10] I think I'm pretty optimistic

[00:59:11] in general.

[00:59:11] I think that's actually

[00:59:12] a good thing, right?

[00:59:13] I don't want to sit here

[00:59:13] mapping Q6s.

[00:59:14] I don't want to sit here

[00:59:15] building security masters again.

[00:59:16] I don't want to sit here

[00:59:17] proofreading a PowerPoint.

[00:59:18] I want to sit here

[00:59:19] like exercising

[00:59:20] high-level thought

[00:59:21] and utilizing empathy

[00:59:23] and social skills

[00:59:24] to talk with other humans.

[00:59:25] So I think in many ways

[00:59:26] it's a positive development

[00:59:27] and you think about

[00:59:28] the history of

[00:59:29] like the labor markets.

[00:59:31] You know,

[00:59:31] over the past 200 years

[00:59:32] there's been massive

[00:59:33] technological change.

[00:59:35] We went from being

[00:59:35] 90% agrarian

[00:59:36] to what, 2%?

[00:59:38] Now,

[00:59:39] and despite that,

[00:59:40] the employment rate

[00:59:40] has basically been the same,

[00:59:41] right?

[00:59:42] Like there's all these

[00:59:43] new jobs that have been created

[00:59:44] because of technology.

[00:59:45] Pilots, right?

[00:59:46] Flight attendants,

[00:59:48] you know,

[00:59:48] prompt engineers.

[00:59:51] And throughout this period,

[00:59:52] right,

[00:59:52] we've seen just a massive

[00:59:53] increase in the wealth

[00:59:54] of society.

[00:59:55] So I think like,

[00:59:56] you know,

[00:59:57] yes,

[00:59:57] the jobs will be different

[00:59:58] moving forward

[00:59:59] but like,

[01:00:00] you know,

[01:00:00] we'll all be probably

[01:00:01] better off for AI

[01:00:02] and,

[01:00:03] you know,

[01:00:03] at least for those of us

[01:00:04] who are,

[01:00:05] you know,

[01:00:06] thoughtful about

[01:00:06] comparative advantage,

[01:00:07] thoughtful about exercising,

[01:00:08] you know,

[01:00:09] development and training

[01:00:10] our ability

[01:00:10] in areas that AIs

[01:00:11] are less competitive.

[01:00:13] I think,

[01:00:13] you know,

[01:00:13] we're actually,

[01:00:14] you know,

[01:00:15] going to have a better time,

[01:00:16] you know,

[01:00:16] doing kind of more

[01:00:18] meaningful work

[01:00:18] than in the past.

[01:00:21] Good stuff.

[01:00:22] Thank you very much, Kai.

[01:00:23] We always,

[01:00:23] these are great conversations

[01:00:24] so please come on again soon.

[01:00:26] Thanks,

[01:00:27] yeah,

[01:00:27] I love coming on.

[01:00:28] Thanks for having me.

[01:00:29] This is Justin again.

[01:00:30] Thanks so much

[01:00:30] for tuning in

[01:00:31] to this episode

[01:00:32] of XS Returns.

[01:00:33] You can follow Jack

[01:00:34] on Twitter

[01:00:35] at Practical Quant

[01:00:36] and follow me

[01:00:37] on Twitter

[01:00:40] and follow me

[01:00:43] on Twitter

[01:00:44] and follow me

[01:00:45] or leave a review

[01:00:46] or a comment.

[01:00:47] We appreciate it.