Weekend thoughts: pattern recognition follow up and AI risk
Last weekend, I put up a post on pattern recognition and used a long example of changes to earnings date as a signal that something was up.
I really enjoyed writing that post (and, judging by the response I got, most readers did too!)…. but it’s not where I wanted to go with the post! So, today, I wanted to “finish my thought” on pattern recognition by taking the post where I wanted to go.
I’d encourage you to read last week’s thoughts for some more explanation / background, but my basic thought was that big money is made in pattern recognition. For example, most event driven investor I know like to spend a lot of time looking at spin offs; at its core, that’s because there’s pattern recognition that spin-offs can be inefficient with a management team incentivized to create value.
However, my big worry with this type of “simple” pattern recognition is that it’s easily replicable. If spin-offs tend to out perform, you can just build a computer model that automatically buys spin-offs. Ditto low P/E stocks, high dividend yields, stocks with insider buying, etc.
Where I think pattern recognition gets more interesting / potentially lucrative is when you’re using it to connect two seemingly unrelated events to see opportunity / risk.
Let me give you a current example that I think fits the pattern (pun intended) / illustrates this point nicely. In the 60s, a lot of conglomerates discovered an “infinite money hack”. Their stocks traded on earnings growth; the faster they grew, the higher the multiple they got. With their stocks at 20x earnings, they could grow by issuing shares to acquire companies at 10x earnings. This growth pushed their multiple to 30x earnings, which let them accretively buy more companies for 10x or even 15x earnings, thus pushing their multiple even higher….
With the benefit of hindsight, this sounds extremely silly. Of course a company can’t create value simply by issuing stock at 20x earnings to buy anything trading at 10x earnings! But, for a while, it worked, and a bunch of conglomerates (including Teledyne) rode that wave and gobbled up everything in sight.
It seems every cycle we learn that “infinite money” hack eventually stops. I remember in the early 2010s there were a lot of yieldcos that traded purely on dividend growth. With their stocks trading at ~20x earnings and private market assets trading for ~10x, for a while the yieldcos had an infinite money hack: issue shares to buy stuff cheaper than your stock, raise the dividend, watch your stock go up, rinse and repeat.
Today, it seems like MSTR1 is following a similar playbook; with their stock trading at a huge premium to NAV, MSTR has an “infinite money hack” where they can issue stock to buy bitcoin and thus increase their bitcoin per share…. plus, they issue so much stock that their aggressive buying probably supports the price of bitcoin.
That hack works great on the way up (and particularly when copycats are aggressively following you into the trade)… but, like all money hacks, I suspect it ends eventually. The question is if MSTR (or any of the copycats) can pull a Teledyne and aggressively repurchase shares if and when the cycle turns and they trade for a discount to NAV….
If I told you, “MSTR today reminds me of the conglomorates in the 60s”, it probably wouldn’t make sense at first…. but I think once you understand the connection there’s a powerful lesson to be learned, and that’s a (simple) example of the type of pattern recognition that I think is really rewarding.
Now, that particular connection might not be a good trade since it’s really about identifying something that’s likely overvalued, and things that are overvalued can get much, much more overvalued before coming down to earth. However, I will note it’s still a useful connection: I’ve seen lots of people avoid blow ups by seeing something that’s trading cheap and making a similar connection to something in a different industry that blew up!
But you can imagine similar situations where a pattern recognition across industries results in huge profits / undervalued securities. Again, a simple example might be best: in general, industries get starved of capital after a big bust, which tends to produce tomorrow’s super normal profits. So if you have bought, say, homebuilders right after the GFC, or post-bankruptcy energy companies right after the mid-2010s oil price collapse, you probably ended up doing pretty well (and I know a lot of people who applied a similar mindset to coal companies and NYC real estate in 2021 / 2023, though its still early-ish on the later). Yes, I’m simplifying / know all of these have idiosyncratic things about them, but I think the general pattern still stands.
So that’s what patterns are and why I’ve been thinking about them recently. Here’s what I’ve been thinking about:
Using patterns to frame investments / research: In the stuff I’m long or looking at, what patterns would apply to the investment that would help me gain more confidence in the investment or exit the investment if I’m overlooking something?
AI risk: obviously AI has been a focus of mine recently. The low hanging pattern fruit (like buying high dividend stocks) has generally been picked over by quantitative models and the like. Right now, there’s still some art to investing that AI can’t recapture (how would an AI model replicate a 60s conglomerate to MSTR and bitcoin?)…. but that will change eventually. How long till AI can instantly analyze a market and tell you every possible pattern a company / market could be following today?
Man with hammer: There’s the old parable “to a man with a hammer, every problem is a nail” (or something like that). Patterns can be insanely useful…. but I always wonder how to balance using them versus applying them to everything. One thing you’ll see happen to a lot of people who make a big market call and get it very correct is they chase the same market call for the rest of their careers. So, for example, most of the investors who nailed the GFC and shorted real estate or bought CDSs (or insert whatever someone who nailed the GFC did) have not been successful post-GFC because you’ll see them chasing the same type of investments / pattern for the rest of their careers (i.e. they bought puts and called a crash in 2007, and then in the 2010s they bought puts and called a crash because the U.S. debt got downgraded…. and then again because they thought munis were in trouble…. and then again when they thought multiples were too high…. and then again because the weather was bad today). How do you use patterns / model as a useful mental tool without letting them become all consuming on one side or the other?
Anyway, that’s what I’m thinking about this weekend. As always, I’d love to hear it if you have feedback / thoughts!
Disclosure: I have a small position