Following the Money: A Capital Cycle Screen in Three Industries

Capital Returns: Investing Through the Capital Cycle is a collection of Marathon Asset Management’s investor letters, edited by Edward Chancellor. I’d call it one of the best investment books of the past 20 years even if you stopped reading just after the introduction.

The key insight is to refocus away from where most analysts are digging (forecasting demand) and instead look at where industries are deploying capital (expanding supply). The supply side has more visibility from financial statements, but oddly tends to get less attention.

That inversion is a useful mental model to identify where future returns might be high, or low.

The Capital Cycle

Here’s the mechanism: Suppose a highly profitable industry (e.g. homebuilding in the early 2000s; software in the 2010s). Managers notice windfall returns being made, and so do investors. Capital floods in, new production is built, and a few years later there is too much capacity. Prices fall, returns fall, and the weaker players go bust or are bought out. Eventually the survivors find themselves in a tidier industry with fewer competitors, and returns recover.

Economists describe this as reversion to the mean in returns on capital, driven by entry and exit. The investing insight is about timing: you want to own the industry when capital is leaving and the market is still gloomy, not when it’s arriving and everyone is excited.

Marathon looked for two kinds of opportunities:

• Candidate A: high returns that, for some reason, aren’t attracting new competition. Barriers to entry, consolidation, or rational competitors keep the excess returns in place.

• Candidate B: low returns with capital exiting. Closures, bankruptcies and consolidation are under way, and returns should improve once supply shrinks.

The book is also candid about where the method breaks:

1. Disruption. If demand falls faster than capacity exits, the cycle never turns. Marathon lost money on HMV, Kodak, Blockbuster and EMI this way.

2. Policy. Chapter 11 lets failed companies keep operating with cleaned-up balance sheets. Marathon calls it “survival of the unfittest.”

3. Cheap money. Low rates keep zombie firms alive.

4. Globalization. A regional industry can become global due to overseas competition. A decline in domestic US capacity may be meaningless if China is still ramping up.

Management is the other half of the framework. Over a decade, a CEO allocates most of the capital in the business, so it matters whether their decisions are just going with the flow of the rest of the industry (expanding when everyone else does) or acting intelligently, in a counter-cyclical manner.

Chancellor’s book was written in 2015. How might we implement Marathon’s strategy, looking back a decade later?

Step One: The Cross-Sectional Screen

I started with Aswath Damodaran’s industry datasets, which he publishes every January from NYU Stern. They cover about 80 US non-financial industries. (I exclude banks, insurers, asset managers and REITs, since their definitions of capital can be idiosyncratic and hard to compare with others.)

For each industry, I built a supply score from three measures:

• Capex relative to depreciation and amortization. A ratio above 1x means the industry is growing its asset base. Below 1x means it’s shrinking.

• Net reinvestment as a share of invested capital. This is capex less depreciation, plus acquisitions and net R&D, as a rough proxy for asset growth.

• Net investor flows. Equity and debt raised, less dividends paid. Is the industry taking money from investors or handing it back?

I then overlaid return on capital less the cost of capital, along with valuation.

Source: Damodaran Online, January 2026 datasets (fiscal 2025 data). Industry ROIC is his “ROC” from the EVA dataset.

The first result was mostly noise. The most “capital-starved” industries were largely in secular decline (the “Kodak camera” problem), or they were accounting artifacts. Industries that make lots of acquisitions carry heavy amortization, which pushes capex/D&A down mechanically. Homebuilders hold their capital in land, which doesn’t show up as capex. And US-only data misses foreign supply, which matters a lot in semiconductors.

Comparing industries against each other mostly shows which industries are structurally capital-light. It doesn’t tell you which ones are retrenching.

Step Two: Each Industry Against Its Own History

The better question is whether an industry is investing less than it normally does. That needs history, so I went back through Damodaran’s archives to fiscal 2006.

Two data problems came up first. In his fiscal 2013-2015 files, about half the industries show depreciation figures that are clearly wrong; semiconductors drop from $19.9 billion to $0.9 billion in a single year. I dropped those industry-years. He also changed his industry classification in 2013, so I mapped the old names to the new ones where there was a clear successor and flagged the rest.

As a sanity check, the data should reproduce the cycles the book describes, and it does. Integrated oil and mining both show an elevated 2007-2012 period followed by a bust.

Here are the industries where capital looked to be leaving on the 20-year view:

Source: Damodaran Online, capex and EVA datasets, fiscal 2006-2025. Medians cover 2006-2024; wireless history starts in 2013.

It’s also worth seeing where capital is pouring in. Software is at 2.45x against a 20-year median of 0.93x. Semiconductor equipment and electric utilities are at 20-year highs, and you can probably guess why. (This is the AI build-out, measured.)

My favorite red flag was basic chemicals. It’s at the 94th percentile of its own capex history while earning a 3.7% return on capital. That is an industry still building capacity at returns below its cost of capital, which is precisely what the book tells you to avoid.

I ranked packaging first: decent returns, consistent data, and real closures under way. Oilfield services came second. Trucking failed a test I’ll explain shortly.

Packaging: The Textbook Case, Mostly

North American containerboard (the brown stuff boxes are made of) looked like the book had been written about it.

Between 2023 and 2025, about 5.5 million tons of capacity closed, roughly 14% of the US total. 2025 alone saw the largest annual reduction on record. Mill operating rates rose from about 92% to 95%, inventories hit a 15-month low, and producers announced three price increases in 2026. Demand, meanwhile, was basically flat; 2025 box shipments were the lowest since 2015.

So returns were improving because supply left, not because demand grew. That’s the capital cycle working as advertised.

The industry also has most of the conditions the book associates with rational competition. Three listed players (International Paper [IP] at about 29% of capacity, Smurfit Westrock [SW] about 20%, Packaging Corporation of America [PCA] about 16%) sell an undifferentiated product priced off a published index, and they announce price increases in public.

However, studying the three companies together turned up things that analyzing each one alone had missed.

The volume plans don’t add up. In a flat market, all three were targeting growth. If they grow as planned, everyone else in the industry has to shrink. When every player targets market share in a flat market, price discipline tends to break.

The better industry structure has been paid for. About $37.6 billion went into the three consolidating deals of 2024-25. By the companies’ own reports, each acquirer is earning less on its deal than on its existing business. International Paper wrote off over half of the goodwill from one acquisition within 11 months.

Discipline looks a lot like collusion. A pending antitrust case (Artuso v. PCA) targets exactly the closures and matched price increases the investment thesis depends on.

Then there’s valuation. PCA, the highest-quality operator, appeared already priced for success. The other two are cheaper but carry enough debt that a serious downturn would likely mean a dividend cut, a downgrade and/or dilution. It’s not a risk I’d like to take, by banking on a recovery in levered cyclicals.

Oilfield Services: Does Capacity Ever Actually Leave?

On the data, oilfield services had the strongest signal of all. Its capex has stayed at or below 1.1x depreciation every year since 2016, after peaking near 2.8x in 2014. After the shale oil capacity boom and bust, their reticence is understandable; we’re now at nine years of not rebuilding.

Big question: does capacity physically leave this industry? When a driller or equipment maker goes bankrupt, the rigs and frac fleets don’t get scrapped. They get bought cheaply by someone else and put back to work. On top of that, the customers are major oil producers, who are sophisticated and bargain hard.

I screened 10 players in the oilfield services space, and while I’m still following the industry, it’s with lower conviction. The things worth tracking are retired versus reactivated fleets and rig scrapping, not just capex ratios. (Trucking fails the same test, by the way. Trucks get resold, and anyone can start a trucking company.)

Telecoms: When the Government Owns the Scarce Input

Wireless looked promising for a different reason. After T-Mobile bought Sprint in 2020 and EchoStar exited in 2025, the US market has three national networks. Few players, high entry barriers and visible pricing are the book’s recipe for rational competition. Network capex was running at just 0.76x depreciation.

Reading the three companies (AT&T, T-Mobile and Verizon) together changed the picture considerably.

Ultimately, the screen was measuring the wrong thing. The scarce input in wireless is spectrum, which carriers buy at government auction, and it doesn’t count as capex. Over ten years, T-Mobile and Verizon each spent almost exactly 5% of revenue on spectrum (4.97% and 4.98%). AT&T ran at 7.4% over the last six years. It works like a toll for staying in the business, and the FCC collects it.

Once spectrum is charged as a normal cost, the industry’s shareholder payout plans exceed what it generates:

Source: company filings aggregated in my own workbooks; spectrum charged at each company’s historical average.

The next big auctions are expected in 2027-28. Unless the payout plans change, they’ll be paid for with debt, on companies already levered at three to four times EBITDA.

AT&T and Verizon spent heavily on spectrum to catch up with T-Mobile’s network, reached parity, and earned close to nothing on that money. Measured from 2023, AT&T’s return on its new capital was 3.2% and Verizon’s was slightly negative. T-Mobile shrank its capital base over the same period and was the one whose profits grew. That is the capital cycle playing out between three companies.

Capacity is also leaking in from the side. Verizon rents its network to cable companies, which added 854,000 mobile lines in a single quarter. Meanwhile all three carriers sell spare wireless capacity as home broadband, competing with cable. And all three are pouring money into fiber at the same time, partly with outside private capital. (A race to build first, financed by other people’s money, rarely ends well.)

The point: wireless isn’t a Candidate A (returns at around 8% aren’t very high) or a clean Candidate B (capital is still flowing in, just not through the capex line). On a spectrum-loaded basis, T-Mobile looks cheapest and is the only one covering its payouts. But I don’t have a finished view on it yet.

What I Learned

Some of this is about the method itself.

• Use each industry’s own history. Cross-industry comparisons mostly surface decline and accounting quirks.

• Capex isn’t the only way capital enters an industry. Acquisitions, spectrum auctions and joint ventures all absorb capital without showing up in the capex line. If the scarce input is bought at auction, add it to capex, or you’ll call a capital-hungry industry constrained.

• Ask whether capacity physically leaves. If bankrupt assets just get resold, the cycle doesn’t clear the way the book describes.

• Read the competitors together. In both packaging and telecom, the most important findings only appeared when the three big competitors were set side by side.

The bigger lesson is a little sobering. In both packaging and wireless, consolidation happened and industry structure really did improve. But the potential gains were largely frittered away, in deal premiums in one case and to the FCC in the other, before shareholders saw much benefit.

For now, packaging is shelved with specific triggers, some oilfield service names are on the watchlist, and telecoms sit on the backburner. Alcoholic beverages, where breweries close and aged spirits inventory gives a visible supply pipeline, are technically on the list from my initial cap-ex screening; in that case the visible risk is declining demand, not supply.

Important Information

Cyclical and commodity-linked industries carry particular risks. Earnings and share prices can fall sharply when prices for their products or inputs reverse, and leverage magnifies those declines. Companies discussed in this article are illustrations of an analytical framework, not recommendations to buy or sell any security. Figures are drawn from public filings and third-party datasets that may contain errors.

Data Limitations: Damodaran’s data covers US-listed companies only, is dominated by the largest firms, and has industry membership that shifts over time. Return on capital is a single-year snapshot, which can mislead in cyclical industries. Data and figures extracted and presented with help from Claude Opus 5.5.

Positions and scope: As of last review date, the author, Nick Bormann, holds none of the positions named here in his personal accounts but may buy or sell any security discussed at any time without notice. Bormann Wealth Management, LLC is a registered investment adviser and holds no positions in the securities discussed here. This article is general research, not advice for your circumstances. See Disclosures and Form ADV Part 2A.