Capital Returns articulates a model of fundamental investing that focuses on two closely-related ideas. First, demand is harder to predict than supply, so rather than driving yourself crazy trying to figure out what consumers will do over the next few months, you should just look at cases where an industry's capacity growth has been below the long-term trend for a long time, and when demand inevitably rebounds they'll mint money before eventually getting around to increasing capex.¹ The related idea is to identify a small but indispensable component of a bigger purchase, and bet that whoever makes it will be able to keep raising prices. If you buy a house in the US, for example, you'll have to pay for title insurance, protecting you against the extremely hypothetical risk that the person who sold you the property did not actually own it. This is a comparatively tiny tax tacked on to a large transaction, that vanishes into a rounding error in a very expensive, often time-sensitive, mostly-financed purchase. If vanishingly few home purchases will fall apart because the title insurance was too expensive, then title insurance companies' pricing power is limited only by the prospect of competition. And, as it turns out, there are basically four title insurance companies, and given how low their loss ratios are, they seem to be pretty cozy in terms of pricing.

These two concepts, the high-margin small component and the laggy capital cycle, actually tie together quite nicely. They're both a way to get some kind of option-like torque out of fairly small changes in end demand. And in some cases, these models bleed together. Sam Zell made a lot of money buying real estate below its replacement value, not because there's some intrinsic anchoring effect that requires a building to be worth more than what it cost to construct—the price of a shack in an abandoned mining town is probably lower than the cost of building a new shack, but you'll have a hard time getting occupancy above zero—but specifically because it implies a gap between current rent and a level of rent that would actually lead to new construction. In that model, rent is a component of the cost of living or working in a given location, but it's a smaller piece of it. If an AI lab has to relocate from San Francisco to, say, Reno, you could argue that the cost of rent was the proximate cause, but that's only a decisive cost when something more important has gone wrong.

There are some companies, like TransDigm, that basically build their business on this. TransDigm identifies component manufacturers who make something indispensable, often for aircraft. They'll buy them, and, typically, raise prices. To a Boeing customer, there's very little difference between a $55m jet and a $55.005m jet, but to TransDigm, there's a big difference between a $1,000 and $6,000 ignition exciter. The underlying demand for commercial aircraft, and thus the underlying demand for air travel, feeds into different end markets with different elasticities, and sellers of the least elastic categories capture a disproportionate share of that upside. This same thinking applies not just to products but to employees. It takes a while to become a licensed nurse, so when there's a spike in demand for them, the market can only clear by raising wages enough that people come out of retirement or work longer. Airline pilots have a similar dynamic, especially since they've been so diligent at lobbying for a 1,500 fight-hour requirement for pilots. Bankruptcy experts are another case, and one that's countercyclical, but very exposed to the political cycle.

Both of these patterns are worth watching for, but there are limits. In the very long run, one thing that crimps demand is pricing, especially the nagging feeling that products in some category are getting expensive faster than they're getting better, for no discernable reason. And on the other side, sometimes demand swings in the opposite direction for hard-to-foreceast reasons. If you were looking at a long-term chart of GPU demand, but you weren't reading papers about scaling laws, you might have looked at someone like CoreWeave as a built-to-fail product that was ramping up capacity and creating an inevitable glut. That might still happen, of course, but that's not how it's played out so far. In the end, one way of looking at these bets is that, while they do lead to higher returns, they're also risk premia: you're getting paid an excess return because there's some risk that the cyclical bottom you think you've spotted is in US-based television manufacturers, or Detroit real estate, or an inventory of eminently-rentable VHS tapes (local minima, ripe for decades of secular decay). "The Cycle" is really an epicycle in an arc where many industries go to zero, and the more your bets are disconnected from underlying demand, the more likely it is that you'll miss an existential industry risk that wipes out the entire market. Life would be too easy if the optimal investment were always a stock screen or SQL query away, but the added risk adds a new dimension: it means that even extrapolating from the past effectively requires you to be a futurist.

The Diff has written about supply constraints, elasticity, and the small-piece-of-the-total-cost approach in a few places:

• We wrote in more detail about the TransDigm model ($).

• Changes in elasticity can have surprising long-term effects: maybe Covid has permanently increased the elasticity of oil consumption ($).

• We’ve considered startups that remix real estate ($).

• A good example of a supply-discipline cycle that made some people a lot of money: the end of two-paper towns gave surviving newspapers temporarily amazing economics ($).

Join the discussion!

1  The book has an incredibly fortunate example of this, noting that it's impossible to predict demand for long-distance flights half a decade in the future, but you can look at Boeing and Airbus' order books and see if we'll have too few or too many planes. Coincidentally, the book came out roughly half a decade before Covid.