The number Buffett actually wants is not in the 10-K: the maintenance-capex problem in owner earnings

Buffett defined owner earnings in his 1986 shareholder letter: reported earnings, plus depreciation, depletion, amortization and other non-cash charges, minus the average annual capitalized expenditures a business needs to maintain its competitive position and unit volume.


That last term is the problem. It does not exist in any filing. Companies report one capex line; the 10-K does not split maintenance capex from growth capex. So every owner-earnings number computed from filings, by hand or by software, substitutes total capex for maintenance capex. That substitution is roughly right for a mature, steady-state business and systematically wrong for anything reinvesting to grow: the more a company spends on growth, the more its owner earnings get understated. Buffett flagged this himself, noting the maintenance-capex figure has to be a guess, and sometimes a hard one.


The screening consequence: on FCF yield or any owner-earnings proxy, reinvestment-heavy compounders look worse than they are, and asset-light mature businesses look best. That bias is one reason mechanical value screens keep serving up melting ice cubes.


Disclosure, and where I ran into this: i am the maker of Moatkeep, a fundamentals site built on SEC EDGAR filing. Writing our glossary entry for owner earnings got me thinking. We publish the formula with the caveat stated on the page: EDGAR can't split maintenance from growth capex, so the metric understates owner earnings for growth companies. Entry here if useful: https://moatkeep.com/glossary/owner_earnings .


Actual question for the sub: how do you estimate maintenance capex in practice? Greenwald's PPE-to-sales method, D&A as a rough floor, management's own maintenance-capex disclosure where it exists, or something else?