VALUE: After Hours (S07 E32): Intangible Value Investing Using AI + NLP with Kai Wu of Sparkline Capital

TRANSCRIPT

Tobias: I hope we’re live. This is Value: After Hours. I am Tobias Carlisle. Joined as always by Jake Taylor. Our special guest today is Kai Wu of Sparkline Capital. How are you, Kai? Welcome to the show.

Kai: I’m good. I’m glad to be on. I like the name Value: After Hours.

Tobias: Well, the idea is that it’s a more relaxed conversation, the kind that you’d have after going to a conference in the bar afterwards, talking about the things that– [crosstalk]

Jake: [crosstalk] real shop.

Tobias: [laughs]

Kai: Yeah, I thought it was After: Market Hours.

Tobias: Yeah. Well, that’s that too.

Kai: I was like, “Wait a second, the market is not closed yet.”

Jake: No, I know that’s it’s confusing. It’s in the middle of the day, but it’s after hours.

Kai: Yeah. It’s 4 o’clock somewhere. [chuckles]

Tobias: Tell us a little bit about Sparkline Capital and your approach to investment.

Kai: Yeah. So, look, I’m glad to be with you, guys, on this value podcast. [chuckles] You may find me to be a bit of a black sheep, but I do actually hail from the school value. I started my career at GMO, actually, and have been a value investor since basically entering the industry over 10 years ago. Worked for Jeremy Grantham at GMO and then was part of a hedge fund spinout there. The number two guy on a quant hedge fund with another GMO partner and then started Sparkline in 2018.

The idea here is to continue to evolve the idea of value investing, but to take into account the changes to the economy that have happened over the past century or so, namely the rise of intangible assets, things like intellectual property, brand equity and human capital that at least based on the data I look at, are comprising an increasing share of the value of companies, especially US, and especially large-cap companies.

The Four Pillars of Intangibles

Tobias: How do you define intangible capital, and how do you then track it and make it comparable on an apples-to-apples basis between companies?

Kai: Yeah. So, the framework I use is I have these four pillars of intangible assets. Because intangibles are by nature heterogeneous. It’s hard to compare apples, in general. So, I think having a framework where you create separate pillars is helpful as a starting point. So, the four pillars are intellectual property. So, that’s like patents, but also trade secrets and any software code, know how that’s held at the corporate level.

Second is brand equity. So, things like Coca Cola, LVMH. Third is human capital. And then, the final piece is network effects. That’s think of like Google search, or Facebook, or even AT&T or New York Stock Exchange, anything where adding more nodes to the network increases the value for all users. That’s becoming increasing large component of value for the big tech platforms, of course.

GAAP vs. Intangibles

Jake: So, then, what kind of adjustments to the accounting? Like, where is GAAP wrong now in the new world?

Kai: [chuckles] Yeah, yeah. So, you’re bringing up an interesting starting point. So, when I first started doing my work on intangible research, I did start with the accounting statements. So, go to 10Ks and 10Qs. One of the inconsistencies that you, of course, are aware of, is that physical capital expenditures are capitalized, whereas R&D and marketing intangible investment is expensed. And so, that’s effectively punitive towards companies that conduct homegrown R&D, homegrown marketing.

Now, there’s a weird thing where if you do an acquisition, you get this goodwill item, you can capitalize that with impairment things. But for the most part, companies that are intangible intensive are penalized relative to those that are more asset heavy, more traditional businesses.

So, the first thing I did was following the work of several academics as well. So, this is well known. I did work on trying to make intangible capital expenditures on a consistent playing field. And so, you can actually build up– you can actually capitalize R&D, say each year and amortize it over some period to build up a balance sheet item, which is intuitive, because now, Coca Cola or any tech company whose book value is 1% of their total value, that becomes somewhat corrected. Where now you’re giving credit to say Coca Cola for having an asset on balance sheet that is the brand that is being created over $100 billion of advertising, or for a tech company, all the R&D spent building out whatever software they have.

I found that to be helpful. So, if you think of your baseline as being say Russell 1000 value versus the market and you say, “What if I redefine the book value component of Russell 1000 value by adding the capitalized intangibles?” Well, the performance has been less bad relative to the market over the past decade, but it’s still not amazing. And so, that led me to the next point where I said, “Maybe we need to consider metrics beyond accounting information.”

I think accounting information has two limitations. First, is that just the amount of disclosures mandated by the SEC are limited. So, you have headcount on the human capital front. But that’s not obviously distinguishing between the CEO and their assistant. So, there’s that component. And then, there’s the other piece just like the input, output function of say R&D or even marketing is pretty nonlinear, where two companies can put $100 million into developing a drug and one ends up with Ozempic and the other with nothing. Or, two companies can buy super bowl ads and one go viral and the other one doesn’t.

So, you really need to focus less on historical cost, which is by definition what accounting does, and instead on the output of the production function that is R&D and marketing. That is what led me to– [crosstalk]

Jake: So, what does that show up then on– Does that just show up as revenue or something?

Kai: No. So, ultimately, yes. Ultimately, the end goal is to model something that will eventually trickle into revenue. One thing I found is, and this is a separate result, is that intangible investment and I’ll tell you how to define that in a second. It tends to lead to profits and earnings per share growth with a pretty long lag.

Jake: yeah.

Kai: If you sort companies by their intangible value and then look at how that filters into EPS over the next decade, the first year actually decreases EPS, because you’re obviously spending money to make money, and then it J-curves up such that in 10 years, it’s a very valuable thing. You’ve now created the moat. You have a patent. You have a trademark of some brand that you can now charge premium prices for the product. But it takes some time for that to filter through to earnings.

Now, how do I define it? That’s where I guess I’m a little bit different than most folks in the traditional value school, which is my background is as a quant investor. I worked with GMO and I got pretty into machine learning and alternative data, I guess relatively early on and ultimately what would become AI, so natural language processing.

And so, a big component is to try to identify unstructured data sources, alternative data sources, to quantify each of the different pillars. An example might be to look at patents is an intuitive place to look for IP or trademarks for brand, social media for brand. Other examples might be LinkedIn posts. Sorry, LinkedIn bios. So, you can build up a snapshot of the human capital stock of a given company at each point in time.

And so, intuitively, these sources have a lot of potential value for quantifying intangibles. And of course, the challenge then becomes just the size of the scale, and the fact that much of this data tends to be unstructured and trying to figure out how to work with that to build structured factors that can then be put into a more traditional quant process which would be effectively an ensemble approach similar to combining a composite of price to book, price to sales, price to earnings, etc.

Jake: Any surprising correlations that came out of the machine learning process? Like, things that you are counterintuitive?

Kai: No. Actually, I think the results are pretty intuitive. If you look at say a list of the companies or the sectors, say with the most intellectual property intensive sectors, it’s technology, communications, on healthcare. On the brand side, it’s consumer discretionary, consumer staples. On human capital, it’s finance, tech, healthcare. The network effects would be communications would be the highest, followed by some tech stocks too.

So, I think it tends to be somewhat intuitive actually the way it all shakes. I think what’s interesting, is that we’ve managed to find a way to quantify things in a systematic way, because most managers will say, “Yeah, of course, intangible assets matter.” Don’t tell me they don’t. [chuckles] Everyone knows that. It’s just a question of how you go about building up a systematic scoring process to identify these things in a quantitative way.

Amazon and the J-Curve

Tobias: Why wouldn’t the value of intangible assets just turn up in greater earnings or cash flow? Or, why can’t you measure it just as a flow rather than having to do these additional exercises to show where that intentional?

Jake: You’re trying to get out in front of it, aren’t you, a little bit and see when is you going to get those big paydays?

Kai: Yeah, I think it goes to the J-curve idea, which is that a lot of intangible investors. You think about Amazon being the perfect example, a company that never made a dime for their investors for decades and now is worth multiple trillions of dollars. How do they do it? Well, they basically reinvested into the business, building out customer goodwill through charging prices that were very competitive for their building logistics networks, the network effects and then ultimately AWS, which has been a big profit center.

And so, yeah, if you look at the trailing earnings, trailing free cash flows, it looks okay, but still very expensive. If you look at it on an intangible investment basis, actually, they’re making really interesting investments into these intangible assets. And for a long time, it would have been a very attractive stock on that basis.

Tobias: So, you say that it reconstitutes these value indexes, so that you find inexpensive relative to some measure makes them– You’re pumping up the measures of some of these, so you can see how truly undervalued they are.

Kai: Yeah, relative to like a more traditional price to book or even a price to earnings, any traditional value metric, you’re going to end up with a portfolio composition that is a little bit different. So, less banks, less insurance, less asset heavy businesses, materials, energy, etc., and more technology, consumer brands, communications, healthcare.

It will vary by region. One thing we find is applying the same methodology to the US into the non-US geographies, you end up with a lot more tech in the US, because that’s where all the talent and innovation is. A lot more brand outside the US, which is if you’re thinking about Europe, that makes sense.

And then, through time as well. Through time, things have changed. Obviously, network effects are a lot more important today than they were 20 years ago, because now you have six, seven billion people connected on Facebook. Back in the day, we didn’t have that. So, these things can change through time.

Also, if you go back 100 years, none of this is important [chuckles] or very insignificant. And so, it is a dynamic process by which intangible assets have become increasingly important for the day-to-day investor.

PhDs vs. MBAs

Tobias: I was looking through your documents and I saw you tracked the number of PhDs.

Jake: Is that a leading or is that a– You want to talk about a PhD? [chuckles]

Kai: Let me tell you a fun story. So, actually, PhDs, at least historically have been a valuable indicator. It’s not just a tech proxy. It’s like adjusting for sector. Let’s say, the companies that are investing more in your high-end talent, folks who are coming from MIT, PhDs and machine learning, those guys are pretty good and pretty sought after. So, the ability to pull those people is an attractive thing. One fun thing I did was [chuckles] I also look at MBAs. So, MBAs are actually a negative signal.

Jake: Oh, damn it.

[laughter]

Jake: I knew I wasting my time. [laughs]

Kai: Yeah. You’re setting yourself up for that one. [laughs]

Jake: Yeah. Stepped right into it.

Tobias: So, there is a positive correlation, the more PhDs you have, the better your returns?

Kai: Stock returns adjusted for price. So, everything’s normalized by market cap. So, in this case, it’s not just number of PhDs, because you’re just getting large-cap stocks. You’re saying, “Hey, per dollar invested, how much intangible capital am I getting out?” Whether it’s PhDs or trademarks or patents, the whole idea is a dividend yield because we’re value investors.

If you build a portfolio simply of the most intangible intensive companies, highest R&D divided by sales, highest PhDs divided by total employees, you basically get tech, [chuckles] which is fine. But I don’t know to what extent that’s persistent moving forward. Because while tech has outperformed historically, right now it’s priced in. Everyone knows that these are good companies. While network effects were, for example, underappreciated for a long time, people know that these things are not important now. And so, now, the key is finding undervalued network effects where it’s not already priced into the stock.

Jake: It would seem like also, if you’re having to fish for these people with $100 million signing bonuses at some point, that’s neutralized a little bit, isn’t it?

Kai: Exactly. Right. [chuckles] If $100 million is coming out of the stock in stock-based comp or whatever, somehow then that’s going to offset whatever value these people bring to the table, for sure.

Tobias: Can you take us through some of the positions, and how they would be turning up in your portfolio and defying more traditional classification of those positions?

Value Investor Buying Big Tech

Jake: Yeah. What’s the biggest surprise jump from what we’d say traditional value would say it was wildly overpriced, or not good or something versus the insight that you have from your process?

Kai: Yeah. Toby and I were talking a little bit about this before the call. I think Mag-7, that’s basically the entire game these days. All people think about is Mag-7. It has, to be fair, driven the returns of the market for the past few years. And so, actually, in 2021, when we launched our product, some of the largest holdings– This is something that’s public, so you can look it up. We’re actually Mag-7 stocks. The Wall Street Journal actually wrote an article on us saying, “The value investor buying big tech stocks,” as if it were a weird thing, which it is. But maybe today, Meta is in the index, but it wasn’t always like that. And the question is, why?

These stocks, many of them traded at price to earnings ratios that were very high. Like, Nvidia was at $100, Amazon was really high. Why would we buy these things? I think the reason is, is that if you add into the mix, these intangible assets, Nvidia has CUDA network effects. They have a culture that’s well known to be innovative. They do hire a lot of PhDs and employ a lot of very talented folks. You tally all these things up. Actually, on a real basis, these stocks were at the time at least amongst the cheapest things on the market.

Now, obviously, fast forward four years and times have changed and everyone knows Nvidia is a great stock and the stock price is up a million times. So, maybe it’s no longer cheap. But you get the idea that when you adjust for these things, it does change the relative. It reshuffles the deck a little bit generally in favor of intangible intensive industries like technology. But more importantly, even within those industries, certain names tend to benefit more from these adjustments than others, of course, as you would expect.

Tobias: So, traditionally, a tech company, they expense whatever they or it runs through a salary or something like that, so it’s not capitalized and depreciated subsequently. It’s expensed in the year that it’s incurred, so that reduces their earnings and their book value as well? And so, then if you make these adjustments, you increase their book value and you’re reducing their earnings– you increasing their earnings as well?

Jake: Increase earnings and then reduce return on invested capital typically.

Tobias: Yeah. That was where I was going. So, doesn’t it make them look less valuable if you do it that way?

Kai: It makes them less attractive on an ROIC or ROE basis. I think Mauboussin done some good work on this, where he looks at Microsoft, and it has an insane ROE. And he’s like, “All right, this thing has insane–” But the capital base that Microsoft has is not true. It’s an illusion. If you adjust for the intangible capital, all the R&D Microsoft has done. All right, well, now the ROE of Microsoft is like a normal level. So, you see that effect.

I think it’s important to also mention that for this, this can actually operate in reverse. So, what really matters is for the earnings-based approach, is the rate at which you’re investing in R&D relative to the depreciation or amortization in this case of R&D. So, in the case of you’re a steady state company, stable company and you’re no longer investing in R&D, you’re just like coasting, you have a lot of R&D capital from prior years is now depreciating or now amortizing, this actually will be the opposite effect.

So, we’re really only talking about companies that are investing in growth and are on this accelerating growth curve, not those that are maybe R&D intensive but are on the decline mature companies.

Jake: You can sweat the assets in a more traditional sense of underinvesting and maintenance and you’re just boosting cash flow short-term. Do you see the same thing when it comes to IP or intangibles where some companies are sweating their intangible assets?

Kai: Yeah. But you definitely want to see companies that are continuing to invest in the future. I think that’s something you do want to see. One interesting thing I’ve been looking at more recently is this, which is– I’m obviously a fan of asset-light businesses, like Buffett and most people. [chuckles] But one of the interesting we think we’ve seen with the Mag-7 in particular has been this pivot from this beautiful asset-light business model like Google Search to what’s effectively like a utility.

Meta is spending 30% of their revenues on capital expenditures, which is the same as the average utility. They have a tech multiple, but they’re basically trying to build utility. There’s many game theoretical reasons why this is happening, and we can go back to prior investment booms and it makes sense that this is happening. It’s not surprising.

But one thing I did look at in the context of this trying to understand the implications of CapEx, is to what extent does changes investment rates lead a presage outperformance or underperformance for a stock. And so, this is a really well-known finding from Fama and French, the so-called investment factor, which is this idea that companies that invest tend to over extrapolate the prospects of the future and tend to then underperform.

So, I did something interesting which I actually bifurcated that into two components. What if companies that invest heavily, but they invest in intangible assets versus tangible assets? The companies that are investing in intangible assets do just fine. It’s really the companies that invest in physical capital expenditures that have historically at least suffered. Again, on average, I’m not saying that this is going to happen to Meta, I’m just saying that on average, this has been the pattern that investment in physical infrastructure has not been rewarded by the stock price.

This is, of course, what we saw in the dot-com boom when you had companies like Global Crossing basically go to zero. We saw it in the railroad boom as well. Now, in this case, it’s a little bit different. I don’t think Meta has any balance sheet issues or any shortfall of free cash flow, but definitely, it’s a concerning development, especially for a component of the market that’s now 40% of the index.

CapEx, AI, and the Prisoner’s Dilemma

Tobias: Yeah, that was where my question was going. I saw a chart on Twitter last week, at the end of last week, that said that the Mag-7 or whatever, the AI CapEx spenders are classified under– They’re spending something like 60% of cash flows, which is double what it was 10 years ago on this build out. And then, the impact of that will be in about three years’ time that their return on invested capital is probably half where it is now as a result of those depreciation, amortization starting to hit the financial statements. What do you think about the likelihood of that?

Kai: Yeah. We don’t really know whether return on this investment will be. There is a state of the world where we achieve artificial general intelligence, and there’s only a few companies that have the service to do so, and they are a cartel and they don’t cut prices so they collectively earn all the rents. There’s a state of the world that that happens.

There’s also one in which artificial intelligence disappoints that it’s not adopted by the enterprise. There’re no real use cases. It’s too faulty. Even if it does succeed in the long run, there’s a state of the world where there’s a short-term hiccup in which case these stocks which are priced close to perfection will rewrite and these companies will have to readjust their investment plans.

And then, there’s the base case. I think, Toby, what you’re outlining is the base case that we’re transitioning from– Think of, I don’t know, Meta as being the average of this amazing asset-light business model and this okay utility model. That’s my base case, which is that they’re just averaging into a less good business. If they could spin out this data center thing, it would be less attractive. But there’s of course these really extreme tails that’s difficult of course to handicap that probability. So, it’s hard to say.

The way I think about things is just from a risk standpoint, which is like, there’s basically been one risk vector, so to speak, that is now driving markets up, markets down and that is Mag-7 and more particular AI. The AI theme within Mag-7. If people are bullish, the stocks go up. And if they’re bearish, stocks go down. It’s not a great market structure, it’s a fragile market where there’s one principal component driving everything.

And so, to the extent, you are an investor, you’re thinking, maybe I’ve done pretty well holding S&P 500, NASDAQ, QQQ, whatever, that’s been great. But at what point do I want to think about taking these chips off and maybe diversifying into other things? How do I do that in a way? That’s not just buying companies that are oppositional to this. They’re not negatively exposed to innovation, they’re just at least neutral or still positive, but maybe just less concentrated in a few names.

Tobias: I think it’s funny to look at the reaction to the market, to the spending now versus when Meta was doing it a few years ago under the Metaverse name, which people just hated and it sold off. I guess that some of that reinvestment was happening before the– It wasn’t CaPex, it was happening sometime further up the financial statement, so it was crushing their free cash flow as well.

So, I guess there were two elements to it, but I feel like there was this fear that everybody had sealed themselves into the rocket ship with Zuck for the run to the Metaverse, and then halfway through two years and he decided he didn’t want to do it anymore.

Kai: Yeah, I remember that when he called off the investment splurge call for the year of efficiency, stock did really well. That would be the dream state for these companies, I guess. That’s the other dream scenario where it’s one in which they all collectively say, “You know what? This was a fever dream, let’s just step away, no more investment.”

One of the concerns I think, is that a large component of the reason why these companies are investing so much is this game theoretical arms race component. Whereas if you’re Google, you need to protect search at all costs, that’s your entire business. And to the extent, there’s something that can potentially impinge on that, it’s not a choice. I think Zuck has even talked about this specifically, but how this FOMO trade and you have to– The downsides of not investing are too high. So, you have this from a personal dilemma standpoint, you’re better off all doing one thing, but they do the opposite because they don’t want one guy to be the one who raises the head like China or SoftBank, and then they miss out. So, they all have to invest, which it makes sense– It’s rational for them to do it, but it’s like not necessarily the collectively the best thing if you’re a shareholder in all those companies collectively.

Jake: No. They had their own little fiefdoms carved up where they were all making a lot of money, but now they’re all going to compete in the same space. Maybe not as much profit pool thereafter when everybody’s in this blood red ocean together.

Kai: That’s right. Yeah.

Tobias: I think it’s funny to see how much internet traffic seems to have fallen off the most big websites, which is clearly people not doing that traditional thing where they had go to Google, go to search, click the link, read the page and now, they use basically AI just gives you a summary of what the top something websites say and say.

Jake: It’s informational gathering?

Tobias: Yeah. And so, a lot of people use ChatGPT or whatever their provider is as their search window, and so there’s no need to go through Google anymore. So, I imagine that Google has seen that from their side as well.

Kai: Yeah.

Jake: When you Google something, are you just reading the top little Gemini prepared section now?

Tobias: Yeah, regularly. I do that.

Kai: Yeah. It just depends on what specifically you’re trying to accomplish. But yeah, I think there’s a significant percentage of search traffic that can be just satisfied at that stage, like you don’t need to even ask further questions or dig any deeper.

Tobias: It’s going to upend the business model for a lot of content providers out there. It’s a very different approach. I don’t know how they deal with it. I’ve certainly seen across a very large number of large websites like the traffic is down, significant numbers in the last few years like 25%, 30% which would be very, very concerning if I was running one of those.

Kai: I saw that Substack is apparently doing pretty okay.

Tobias: Substack was the only one that was up and still performing really well. Yeah, we probably saw the same chart. Yeah, that’s right. Substack was the top left-hand corner.

Kai: Yeah. Why do you think that is?

Jake: Left corner though. That’s the real question.

[laughter]

Kai: I’m assuming they’re blocked on ChatGPT.

Tobias: Yeah. I guess they’ve got an email direct to consumer type business model where how often do you go to the substack.com to read it? Often it’s emailed to you and you just read it as an email.

Kai: Again, look, maybe I’m an optimist here and I’m just warping things to my worldview. My thought would be that much of the content is commoditized. They’ve already scraped Wikipedia and just the random web. And so, what really matters now to get the incremental improvements on models are the really truly unique content. And so, long tail stuff, because it’s been shown that it’s not about the amount of content, it’s about the quality. Like, one good data point, one good piece of content is more valuable to training a model than a thousand like whatever, just noisy low quality. So, to the extent that is true, which maybe that’ll be proven wrong.

I think Reddit, for example, is a really important source of training data for a lot of these models, just because it contains, as we discussed this long tail of interesting opinions and whatever. So, you could envision a state of the world where that actually ends up continuing down that path where you have–

Even before AI, you get all these content mails. They’re generally outsourced to other countries where people just put slop onto the internet and use it to generate SEO traffic. Like, that I assume is going to be dead, because now you can just do that on AI. But at some point, well, that’s not really incrementally useful to the society or to the AI models more importantly, and so you’d expect that people who are creating genuine content would actually be advantaged in that world. But I don’t know, maybe that’s just me being optimistic.

Jake: Yeah, how good can AI be if it’s just trained on listicles? [laughs]

Kai: Right. Exactly.

Tobias: Buzzfeed articles.

Jake: Yeah.

Kai: I will say, when I’ve used AI in my writing, I haven’t found it to be that useful. At least, this is as of today that useful and accurately generating the writing. But one thing it’s really good at is generating headlines. I’m pretty bad. I will try to post on LinkedIn or Twitter like, “Oh, yeah, I just put out this new research piece.” I’ll say it in my own voice, and it’s pretty boring. It’s like an academic, whatever.

Jake: [laughs]

Kai: And then, [crosstalk] I’m like, “ChatGPT, make this better.” It’ll give me this really snappy headline, like take out the key stat. Or, 40% of people are– You put it up there, and I suddenly get so many more clicks, so I don’t know. Obviously, it is trained on social media data and it does know what makes a good headline, which I don’t know. So, I guess that just goes back to the general idea, which is that we should be using AI in areas that we’re below average.

Because AI always gives you the average outcome. If you’re really good at something, don’t use AI. If you’re really bad at something, then yeah, you should use AI. I’m really bad at writing headlines, so AI is a great companion for that.

Tobias: It’s got very mid tastes, I find. I always try to get it to do some writing for me and it writes in this really predictable cliched midline way that is just so incredibly boring to read, because it’s just a mishmash of everything you’ve ever read before, which I find really frustrating. But I think you can train it to give you good titles. I think it is quite good at that.

Kai: Yeah. I think to your point about it giving mid-content, I think that has only become worse over time, because a lot of the content we see online, like, maybe we don’t really truly internalize this fact, is actually AI generated now. My guess, if you scroll through Twitter or whatever, you just look through some stuff, that a significant percentage of the content is actually AI generated. And so, it just starts to all blend into the background.

And so, part of what you’re saying is maybe five years ago, you would have been like, “Ah, this is a game changing thing. This content is amazing.” And now, it’s just everything. They all use the em dashes and have the word delve or whatever.

[laughter]

Jake: Yeah.

Kai: Well, this is AI generated.

Tobias: That’s a– [crosstalk] Somehow, I like the em dash.

Jake: I like it too.

Kai: I actually like em dashes. I’ve been using them for years. I’m annoyed that I can’t use them anymore.

Tobias: Yeah. Yeah.

Jake: Yeah.

Tobias: If you write in Google Sheets, which over Google whatever it is, which I prefer because then it’s available from computer to computer, the em dash is just three of the dashes. It’s so easy to put in. It’s harder to do in word. Yeah, now it turns out that’s a sign of AI writing, which is a bummer.

Kai: Right.

Jake: Ruined it for you.

Tobias: JT, top of the hour. You want to take a chop at some veggies?

Jake: I will. I knew that Kai was coming on the show today, and I knew that he was interested in this topic. I saw him post the tweet about it, and I’d been reading it anyway, so I’d been saving it for when he was coming on.

We’re talking about cash holdings. It starts with this question of, why do some companies run their liquidity super tight and others sit on these mountains of cash? And so, you might be thinking like, “Well, isn’t having more cash always a good thing? Isn’t it safer?” As with most things in business, the answer is it depends. It’s complicated. So, cash can be a blessing or a curse in some ways.

So, too little is obvious. The company risks collapse or is fragile to downturns, but too much then, it can risk dragging down returns on equity, obviously, but it can also frustrate investors when it’s sitting there. Probably, most importantly, it’s tempting for company management to do these creative or destructive, often ego driven decisions, M&A activity, when you’re sitting on a fat wallet, you can make some stupid decisions.

So, there’s this tension between safety one side and stagnation the other. That is at the heart of this recent white paper from Michael Mauboussin and Dan Callahan that was called Cash Holdings: Data, Theory, and Alternatives. So, we’re going to do a little mini book report today on that.

So, Michael, he analyzed over five decades of data across industries, lifecycles of businesses, global markets to understand how companies think about cash and how have they expressed that. Their conclusion is that there’s no universal right amount. Instead, cash is a strategic lever to be run inside of a business. So, we’ll start with some of the scale of it.

At the end of 2024, the US public companies, excluding financials, were sitting on about two and a half trillion of cash. That’s no small change. It’s about 4.7% of their total market cap and 9% of their total assets. And for reference, the average from 1970 to 2024 was about 7.5%, so we’re in normal territory right now.

This cash is not evenly spread out across the corporate spectrum. A handful of giants account for a very disproportionate share of it. Just 10 companies, the top 10, hold one quarter of the excess cash. About a third is held by just 21 companies. Entirely half of that two and a half trillion lives only 67 balance sheets. So, [chuckles] it’s a power lot–

Tobias: I thought you were going to say seven balance sheets then, but that’s fine. Keep going.

Jake: [laughs] No. And so, in Berkshire, which was technically excluded from the study because it’s labeled as a financial, they’ve got north of probably $350 billion at this point on their balance sheet. So, since the 1970, companies have steadily raised their cash balances from an average of about 6% of assets in the late 20th century to almost 10% in 2001 to then 11.5% in 2020.

Now, not every industry hoards cash in the same way. Healthcare and technology dominate the rankings. In 2024, the median healthcare company had nearly 40% of its assets in cash. For biotech firms, a lot of times, they’re pre revenue and so they’ll hold even higher, sometimes more than the market cap entirely is held in cash. Technology isn’t far behind them, with the medium at 23%. So, think of software firms with big R&D pipelines.

And then, by contrast on the other side of the world, you have utilities and energy firms, which they rely heavily on tangible assets and they enjoy relatively steadier cash flows often, especially on the utility side. They’ll hold very little cash then, and often less than 4%. And in fact, even sometimes utilities are forced to take on leverage from the ratepayer’s perspective wanting to have a lower cost of capital.

So, the split reveals a deeper story as well. It’s not really just about risk management necessarily. It’s also the nature of the business and the asset the cash has been supporting, so there has to be some common sense to that. If your assets are all intangibles, like codes and patents, algorithms, all the stuff that Kai’s been talking about, those things don’t make for very good collateral.

Banks are pretty hesitant to lend against them which means that companies that are reliant, that are reliant on tangibles, they end up self-insuring that by holding more cash. As you know, over the last few decades, the corporate world has shifted from building factories to building a lot more intellectual property.

So, let’s see. The life cycle of the company is the next thing. This also plays a critical role in cash policies. As you’d expect, early-stage companies, pre-revenue companies, they tend to hold a lot more cash. Their operations and investments, they eat money. Finance can provide a lifeline, but at that stage, you need to hold a lot of cash. The median then firms for early stage is 31% of assets in cash.

At maturity, a company is ideally cash generative and then the ratio then falls to about 8%. And then, in decline, actually cash balances arise again, as sometimes there’s divestitures that a company can do or a shrinking operation that might be in runoff, and so you’ll see cash balances balloon again. So, there’s this U-shaped curve in the life cycle of the business as far as cash balances go.

All right, quick quiz question. Was corporate debt to total capital higher in 1974 or 2024?

Tobias: 1974.

Kai: 1974.

Jake: Well, you, guys, are correct. [Tobias laughs] Of course, you would be. So, 40% debts of total capital in 1974 and 15% in 2024, which is a little bit surprising. You have this narrative that like, “Oh, everyone’s over levered today.” Maybe not true.

Now, why do firms hold cash? Economists and finance scholars point to four explanations. Number one, precautionary motive. It’s the simplest, like cash as an insurance. When markets freeze, companies with cash survive. COVID-19 revealed that in a big way. Airlines, which had been spending their free cash flow on buybacks, all of a sudden faced literally zero revenue, what do you do then? You need a cash cushion.

Number two, tax management. Before the 2017 Trump tax cuts, there were lots of cash that was parked overseas, these US multinationals. They didn’t want to repatriate it and pay the taxes on it. So, that led to these massive offshore cash piles.

Number three is option value. Alice Schroeder had said that Warren Buffett thinks of cash as a call option on every asset with no expiration, which I thought was a pretty eloquent idea. And in volatile markets, ideally, a cash rich company has this antifragility to it where they can strike quickly and buy distressed assets or invest in opportunities that their rivals can’t touch, because they’re flat footed with not having enough cash.

And then, the last one is agency theory. This is probably the most cynical explanation. So, this would be managers may hoard cash, because everyone else is doing it basically in their industry, and they don’t want to look like an idiot and be cut out. So, reducing their own personal risk by doing what everyone else is doing, so maybe that means running it at a red line of cash where you’re really low and levering things up, because all your competitors are.

And then, there’s some studies that show that public companies hold twice as much cash as private ones, which suggests that governance and incentives do play a role in this. So, we’ll wrap this up here. It’s already running a little long. [chuckles]

So, peer averages aren’t a policy. They’re really more of an alibi, which speaks to that, you know, what we’re just saying with these principal agent problems. The real question isn’t how much cash, but really what would make you spend that next dollar or give it back. I think that’s an important part. If the board can’t really answer those cleanly, the market will answer for you.

And then, lastly, cash is a mirror. It shows fear or hubris or readiness to act. So, if you read the balance sheet and the cash flow statements, and you can often see the strategy that’s been revealed, you can look through 10 years’ worth of cash flow statements and get a pretty good sense of what management thinks. Or, sadly, sometimes the lack of a coherent strategy, which also comes up. But when it’s done well, it probably rivals one of the most important levers that you have to pull in the investment equation.

Tobias: Good stuff, JT. Usually, I’d go around the horn and give everybody a shoutout, but my text window is not open today and I can’t get it to open up, so I’m sorry about that. I don’t know if the text is appearing for everybody watching from home or not, so I hope it is. Well, I’ll check it on the recording.

Kai, you did a piece on crypto where you’re talking about some crypto factors. Are there factors in crypto beyond momentum?

Jake: Volume.

Kai: [chuckles] Yeah, I did this paper and I got savaged by the value community.

Tobias: By the value community or the crypto community?

Kai: By the value community-

Jake: Both.

Kai: -by Cliff Asness [crosstalk] Financial Times.

[laughter]

Jake: [crosstalk] Oh, no.

Kai: No. To be fair, look, isn’t this what he does? [chuckles] No, but look, basically the idea of this research was to say, “Okay, we obviously know there’s a market factor in crypto, like there’s in every asset class, that beta is not perfectly correlated to the stocks, and so that’s its own factor.”

Tobias: How do you measure it? Is it bitcoin?

Kai: In this case, I did a market cap weighted index of all crypto, which is basically bitcoin up until recently.

Jake: [laughs]

Kai: And then, there’s a small-cap factor, so small versus large, the same way you define it in Fama-French with stocks. And then, momentum, which you mentioned. And then, the final thing I did and is to create this intangible value factor. So, I mentioned intangible value in stocks, and how that’s created using patents and how much price versus per user in the patent.

In the case of crypto, it’s a little bit different, where the entire value of these companies are generally being built in the open. So, you don’t really have patents, you have open-source code. But what you can do is you can go on the GitHub repositories and say, “Hey, how many developers are working on this? How many different commits do you have? How many lines of code?”

The nice thing is, of course, on the blockchain ledger side is by definition for each chain, you can see in real time basically how much activity, how much dollar transaction volume is flowing through the ecosystem. So, think of it as basically opening visas 10Q every second. You can also look at daily active users, so, a SaaS style metric, how many unique wallets are operating, how actively are they transacting defining activity in different ways?

You can actually create, at least in my mind, fundamental metrics the same way I mentioned across these four pillars, human capital, brand, IP network effects in crypto, now that just the data sources are a little bit different and in some ways, actually better. And so, basically, I built this. Without going too much into details, that was the fourth, I guess, fifth factor or whatever.

So, then I said, “All right, let’s try to go around and test each of these things.” And so, market worked. Small-cap has historically worked a little, not more recently. Again, it’s maybe the same thing that’s happened in stocks where Mag-7 is dominating stocks and maybe bitcoin is dominating until more recently.

Crypto momentum is interesting, because it does seem to work historically. The one thing is that it operates in a faster cycle. So, whereas in equities, you typically look at 1 month to 12-month momentum, so 12 minus 1. In crypto, it’s more a matter of weeks.

Jake: Which makes sense. [laughs]

Kai: Maybe it works in seconds too. Maybe there’s actually an inversion on a second horizon. I bet you there is because of the microstructure. But at least on the weekly horizon, say, one month to one week or six weeks or whatever, to one week, you’re seeing some strong trending behavior. You also find something interesting when you intersect. I guess the fourth one is intangible value. You do find a value factor or this intangible value factor in the data.

So, one interesting thing shows up when you intersect these factors with size. So, think of like Morningstar style boxes. Small-cap value, large-cap value, small-cap momentum, large-cap momentum. In crypto, at least, momentum tends to work better in large-caps. That’s a little bit counterintuitive, because in stocks, it works better in small-caps, or it has historically.

I think that has to do with a few things. There’s the argument that the largest assets are more macro assets these days, like bitcoins, digital gold. It’s a way of getting exposure to fiat or whatever. And so, therefore, that’s where you see a lot of the trends. There’s also a lot more narrative attention is being spent on the large-caps, whereas the small-caps are forgotten.

And then, the final piece just being leverage, that you can lever up bitcoin and eth and these sorts of assets, and so you end up with these cascading effects where you have these virtuous and vicious cycles due to leveraging and deleveraging. You don’t really have that in small-caps, because you can’t really lever up your flower coin or whatever.

Jake: Thank God. [laughs]

Kai: And then, I found that intangible value tends to continue to work pretty well, works better in small-caps than large-caps, which is consistent with the way that we’ve seen things work in equities.

Now, there’s one interesting wrinkle here, which I brought up in the paper, which is one of the things that people talk about, going back to Cliff Asness, is this idea that small-caps have generally not worked because of this junk component, that a lot of small-cap are unprofitable and junky. And so, if you filter that out, it tends to work again. So, I did the same thing where I said–

If you look at intangible value, one thing that’s interesting, is that by definition you’re looking at price to X, where X is some fundamental metric. And so, by definition, anything with a divided by zero goes away. So, you’re removing the junk. It’s almost like a quality filter, I guess, in the small-cap universe. And so, that explains a little bit in my mind at least, why there is less of a small-cap effect today– Lot of these tokens are basically vaporware. There’re no actual users on these things.

They get launched in 2021 in the boom, and then no one uses them anymore and they just sit there as a zombie. In the same way that if you’re a Japanese company and you never got delisted after 1990. You’re just sitting there as a zombie, which is fine, but the problem is that you can never delist. These things are always going to be on the blockchain. By definition, once they are created, they’re there. And so, you need ways to like filter out some of the speculative tokens, the vaporware. And in doing so, you actually can restore a lot of the perceived missing premium.

Tobias: Has anybody else ever done factor work on crypto?

Kai: I saw that Kraken and CF Benchmarks came up with a piece recently. I’m not exactly sure what it said, but I did see that they put something out on that topic. There have been a couple academic papers that I have looked at these sorts of things. I don’t think they’re widely circulated, but you can– I think I cite them in my paper, so they’re out there.

Yeah, my guess is that what we’re seeing in crypto is like what we saw in equities, where it’s like, you start out with like just bitcoin, and then over time you build out these index funds, like Bitwise 10, these sorts of products, and then you start to see smaller cap things get launched and then, eventually, you have asset pricing models that come up in order to help investors understand what’s going on in the long tail of assets.

And that’s when the style premium, basically, AQR built their business. Pick on Cliff on this whole idea within equities. I think it may not be me, but I think somebody will eventually over time build the same edifice out in the crypto space, why wouldn’t they.

Tobias: What was Cliff’s critique of the paper– [crosstalk]

Kai: I don’t think he likes crypto.

Tobias: Okay.

Jake: [chuckles] Just stop there.

Kai: Yeah.

Tobias: Because somebody gets the Nobel Prize in Economics for this. So, it’s you or one of these other guys who publish their paper first? You’re the Fama-French of crypto?

Jake: Yeah. There you go.

Kai: I hope not.

[laughter]

Tobias: Back to the equity side.

Jake: Real world.

Small Caps vs. Large Caps

Tobias: Yeah. The phenomenon that I have observed and I’ve been tweeting out a bit recently, is just this dominance of large-cap over small-cap. It turns up anywhere you want to cut it. Like, if you look at the S&P 100, which is the largest 100 versus the 500. And in this instance, the 500 is the small. You still see the same effect where you get the 100 massively outperforming the 500. You see it in equal weight versus market capitalization weight, you get almost exactly the same shape.

So, does any of the work that you do– Does it put that into any context? Does it explain that market capitalization? Does it normalize for it, or how do you deal with that?

Jake: Toby wants to know when it’s going to stop [crosstalk]

Tobias: That’s exactly what I’m asking.

[laughter]

Tobias: That’s exactly what I’m asking.

Kai: Yes, small-caps have underperformed large-caps. The counter argument to what you said, is that well, actually, large-caps have had earnings growth that has justified their appreciation. The Mag-7 have actually had earnings growth, we’ll see whether it persists, that has been pretty impressive and has justified the increase in their price. Whereas small-caps, international stocks, emerging market stocks, these unfavored categories have not.

I focused more on international stocks. I did a whole paper on this where I found that over the past, what, 10, 15 years in real US dollar terms suggested for inflation, international stocks have basically had no earnings growth, which is pretty disappointing. And to some extent, that’s, I think, at least the reason why they’ve underperformed. So, not only have their earnings been tepid, but investors are struggling moving forward and hence they have experienced no, say, margin valuation expansion.

Whereas in the US, they have, because the index has become “higher quality, more growthy,” and as a result, investors are extrapolating that moving forward. And that same story I believe can be applied to small-caps, which is small-caps companies have underinvested intangible assets. Think of the example of JPMorgan and a lot of these big companies that are hiring all these AI engineers and launching blockchain initiatives. You have the original banks, which I don’t know what they’re doing. They’re not doing well.

So, I think you see this across the economy both in terms of value versus growth, US versus non-US, large-cap versus small-cap, where the folks who have invested in tangible assets over the past two decades have actually experienced higher growth. You can show this.

I mentioned that study I did on a cross section where I said, there’s a J-curve where companies that invest in intangibles have then subsequently gone on to outperform those that have not. And that can be applied on the aggregate level to different groups, whether it’s US versus Japan, or US versus Europe or US large versus US smaller value group versus growth. So, a lot of the perceived discrepancies that we’re seeing in growth rates across groups I think can be explained at least partially if not fully in many cases by this intangible discrepancy.

Now, multiples have also changed. Perhaps, they’re overshooting in some cases. I think in most cases, they are overshooting. And so, in other words, said differently. I actually debated Cliff Asness on this too, is the value growth spread, it seems like at all time wides and will that lead to future stock returns that are extraordinary for the value versus growth. I think there’s two components. One is that if you adjust your intangibles, the gap’s less big, but there’s still a gap. So, maybe it explains half the gap in this example. So, I think that it does help explain what’s going on and primarily through the earnings channel, but that the multiples are still a little bit over extrapolating I think in this case.

Tobias: Is it a compositional problem? Because accounting doesn’t deal with software particularly well, because software is unlike a physical edifice. It’s something that the expense runs through the salaries of the engineers who put it together and there’s not a lot of material that goes into it. And so, that’s where all the problem is. To what extent do you see intangible predicting returns outside of tech?

Kai: Yeah, it’s interesting. So, you can run these intangible value studies both within across the market, where you’ll have to go bottoms up and choose the sectors or you can constrain it to say, “All right, I’m just going to pick the most intangible value stocks within each sector or within each industry or within each country.” And in general, you find that intangible value works in both the new economy and the old economy. So, it works in software obviously, as you might expect, but it also tends to work in even sectors that are more traditionally asset heavy businesses.

I think the reason why is just that there aren’t any growth investors who are even trying to price these things in those sectors. Growth investors are all allover tech, and they just have basically left the old economy for dead. As a result, if no one’s looking at it, even having a small edge, even if the edge is only on a smaller share of the balance sheet, that’s still meaningful.

Now, what’s really interesting, actually I looked at this and I think this is actually worth mentioning is the counter of this, is the opposite of this, which is traditional value. Take the most basic Fama-French press the book vector. If you are careful in how you constrain the universe to only be old economy or to be more precise, low intangible intensivity companies, actually, it’s done just fine. So, traditional value has done just fine if you exclude the sectors of the economy that are the regions, the stocks that are more intangible intensive.

Of course, the problem there and the counter argument to that, the problem with that, is that that share is becoming diminishingly small in the US. That used to be most of the market. And over time, it’s becoming smaller and smaller, because intangible assets matter for every company now, maybe, less for some than others, but is still important. But I think it does at least make a point on principle, whether or not it’s useful in practice that maybe it is the case that some of these factors have been underperforming, because the omission of these intangible components which when added should in theory correct it.

Tobias: It’s interesting, because one of the observations that Cliff made was that price to book did work pretty well through that period of time when value was really getting hammered specifically, because price to book was the less good value factor and so–

Jake: Yeah. The worst value factor. The one least representing value, so it did better. [laughs]

Tobias: Did better. The weakest value factor did the best. Yeah.

Kai: Yeah, I think that speaks to the cyclicality part. These types of value factors, there’s two reasons they outperform. One is just there’s a recycling effect. Value stocks prices rise more than their earnings underperform, and that growth stocks underperform because they grow less than the market’s pricing in. And that effect seems to have diminished, at least based on what I look over time. Whereas the cyclicality component, where there are groups of stocks, tech versus energy, US versus emerging, value stocks defined this way versus growth stocks that do go through cycles. I think it is undeniable that value stocks are in a disfavor sector right now. At some point, that rubber band will likely snap back and these stocks will likely experience a cyclical rebound.

Now, will the trend line continue to be– If you go back to the Fama-French chart since what, 26, there’s just been this nice five percentage point of smooth outperformance for value versus growth. And then, obviously, the past 15 years it’s been falling off that. It may rebound, but will that necessarily mean that the trend line will continue to be up, that I think is more open to debate.

Trade Wars and Intangibles

Tobias: I saw you wrote one of your pieces on trade wars. Do you have any view on the tariffs?

Kai: Yeah. So, I wrote that paper maybe a week after Liberation Day or two weeks after Liberation Day. My point there is a lot of people were starting to get concerned about geopolitical risk. And to the extent that most people of home bias and they were mainly invested in US stocks, or effectively dumping their multinationals to buy domestic stocks. RH is the best example of a stock that went down 50% or something in a single day to go buy utilities and save stocks.

And so, what we wanted to do was ask the data and say, over the long arc of history would actually have been a good strategy. What I found is that multinational stocks, whether importers, exporters or pure multinationals at both import and exports, that that class of companies has massively outperformed their domestic peers, whether in the US, in Europe, whether adjusted for sector, just on all dimensions, that these companies have done better.

One explanation is selection bias, that if you’re an A player, you want to compete on a global arena where the TAM’s bigger. If you are a B player, you want to hide in your local market, and use tariffs and export controls or whatever, import controls to protect your niche. So, that could be one explanation. But it could also be the fact that being able to be connected in a free trade ecosystem is actually helpful. The idea of outsourcing to lower your costs, to access customer bases across the globe. There’s a lot of tax advantage that multinationals can access that those who are stuck in one country don’t necessarily have the ability to do.

So, anyways, I thought that was a really interesting finding. The other finding I homed in on related to integrated intangible assets is this idea that, yeah, well, intangible assets– So, first of all, multinational companies tend to be more intangible. There’s a positive correlation. Not all of them, but sometimes. It’s an imperfect but positive correlation.

But the other thing is that when you think about tariffs, what is tariffs really doing? They’re making it really hard for a good that’s being exported through over a border to pass without a tax. Intangible assets by definition don’t have to pass borders, so you can’t tariff an intangible assets. And so, it became a really interesting.

A way of building a portfolio that is still on globalization and still on free trade, which to me is capitalism’s golden goose. I like free trade. But to do so in a way that is more insulated to some of the risks that we’ve seen with governments being able to capriciously put tariffs on their trading partners.

There’s obviously shades of gray, whereas you have pure intangible, revenues like software licenses, and then you have pure physical goods and then you have the in between which are Apple’s iPhone, which is, of course, a physical asset, but a lot of the value there is the IP. And so, there’s some transfer pricing gets really complex. But there’s, of course in between category as well. But staying the most intangible intensive companies tend to actually be relatively well positioned for trade wars.

Tobias: Yeah. It’s fascinating stuff. We’re coming up on time, Kai. So, if folks want to follow along with what you’re doing or get in touch with you, what’s the best way of going about doing that?

Kai: Oh, you can go to my website. It’s just sparklinecapital.com. You can read my research there. I have a form. You can submit any messages to me or you can email me directly. You can find my email actually on any of my white papers which are on my website. It’ll be in the upper left-hand corner.

Tobias: And you’re on Twitter too, @ckaiwu.

Kai: I am. Yeah, my handle on Twitter and LinkedIn is the same. It’s @ckaiwu. C-K-A-I-W-U.

Tobias: Good stuff. JT, any final words?

Jake: No. Thanks for coming on, Kai. It’s good seeing you.

Kai: Yeah, I had fun. I’m glad I could join Value: After Hours.

Tobias: [chuckles]

Jake: [laughs]

Tobias: Thanks, Kai. We’ll be back, folks, next week with any luck with the chat fixed up again. So, I hope that’s working then.

[Transcript provided by SpeechDocs Podcast Transcription]

VALUE: After Hours (S07 E32): Intangible Value Investing Using AI + NLP with Kai Wu of Sparkline Capital

Read next