intrinsic-value

Compute what a business is worth as a business — owner earnings, return on tangible capital, a reverse DCF, and a margin-of-safety ladder — then rank a set of names against each other. Use for "what's

SKILL.md

GitHub

Intrinsic value — the arithmetic

Setup. The scripts sit beside this file. Run them from a working folder that holds your .env (E*TRADE consumer key/secret) and a data/ directory; install deps with pip install -r requirements.txt where one is included. Part of a set: etrade-pull, order-check, portfolio-review, realized-pnl, rebalance-plan, price-history, trade-journal, intrinsic-value, sec-financials, buffett-checklist.

The theology is: simplify everything. Buffett has never published a DCF. He capitalises owner earnings against the long bond and then demands a margin of safety instead of inventing a risk premium. If the answer needs a nine-tab model to hold, the answer is "too hard" — a legitimate verdict and often the right one.

Deliver one number and the one relationship that produces it.

python3 owner_earnings.py CTVA --price 82.81 --bond 5.19    # one name, full ladder
python3 compare.py --bond 5.19 RMD=232.45 KO=91.99 ...      # rank a set

owner_earnings.py prints owner earnings by year, return on capital, what the market is asking, three value anchors, a margin-of-safety ladder, a reverse DCF and a sensitivity grid. compare.py ranks a set on the expectations gap. Both take every input from sec-financials — nothing is typed by hand.

Owner earnings

Buffett, 1986 Berkshire letter, Appendix:

reported earnings + depreciation, depletion, amortisation and other non-cash charges − the average annual maintenance capex (− additional working capital where the business needs it)

He adds that this is "the relevant item for valuation purposes" and "does not lend itself to the precision of GAAP" — approximately right beats precisely wrong.

The rules — each one earned by a name that broke the model

1. CALIBRATE BEFORE DELIVERING A VERDICT. Non-negotiable. Run the identical model on a business Buffett actually owns before saying a word about the target. Run on Corteva it returned 92% overvalued — and on Coca-Cola, 116% overvalued, against 40.7% ROE and a 72% return on tangible capital. The verdict was never about Corteva; it was a 10% hurdle applied against a 5.19% long bond, which calls the entire market overvalued. An uncalibrated intrinsic-value model is a permanent bear wearing a spreadsheet. Never deliver the absolute number alone.

2. Never value on EBITDA. Convert it, and state the ratio. Munger: "every time you see the word EBITDA, substitute the words 'bullshit earnings.'" Corteva's owner earnings are 44% of its operating EBITDA, so a 13x EBITDA bid that sounds cheap is 29x owner earnings. That one ratio has explained the whole bull/bear gap every time.

3. The reverse DCF is the deliverable; the DCF is scaffolding. Nobody can adjudicate your growth assumption. Invert it: at an 8% required return, what growth does today's price require? Corteva answered +9.2%/yr for ten years from a business whose owner earnings had fallen 8.5%/yr for four. That is a fact about the price, not an opinion about the company.

4. Pick the base by testing the series — this is the easiest way to be badly wrong. Averaging five years is right for a cyclical, whose good years fund its bad ones. It is wrong for a compounder (it averages in years the company has permanently outgrown) and wrong for a spin-off (it averages in carve-out losses belonging to the former parent). Un-fixed, this made NVIDIA look like it needed 25% growth and Constellation 28% — both artefacts. choose_base() tests monotonicity and says which basis it used; read that line every time.

5. Owner earnings must track the cash. If they don't, refuse. A large positive owner-earnings figure against weak or negative free cash flow means the D&A add-back is carrying a goodwill impairment, not depreciation — and adding back a write-off of money already lost values a melting business at a premium. FMC screened at a 57% owner-earnings yield on a −108% ROE; Kraft Heinz at 7.3% on −14%. Both were impairments. The guard is owner earnings ÷ free cash flow > 1.6 → REFUSE.

6. "Too hard" is a valid output — Munger keeps three baskets: in, out, and too hard. Fewer than three positive owner-earnings years in five is not a cheap stock, it is an absence of the thing being valued; averaging the winners is data-mining. Loss-making in the latest year is an automatic refusal. So is a business the formula does not fit: banks, insurers and GSEs have no meaningful capex or D&A — value those on book value and ROE, not here.

7. Never extrapolate a delivered CAGR above ~30%, and check it against forward guidance, not just history. The expectations gap compares the growth a price requires to the growth a business has delivered — which silently assumes the past is the right comparator. It is not, once a business inflects. Lululemon screened as the cheapest name on the board (implied −3.2% vs delivered +14.7%) while management guided EPS down 16% and gross margin fell 550bps. Cheapness does not save a declining business. Before acting on any negative gap, read the latest guidance and the gross-margin direction. That is where filter 2 lives.

8. Split D&A before adding it back. Depreciation ≈ maintenance capex and is a real cost. Amortisation of acquired intangibles is a non-cash residue of a past purchase price. Add it back only when the company separately expenses the R&D that renews the asset — Corteva expenses $1.47B/yr of R&D while amortising $644M of 2017-vintage intangibles, so the add-back is correct, not generous. Where the asset is not separately funded, adding it back is how you talk yourself into a melting asset.

9. "One-time" charged every year is opex. Corteva booked restructuring in four consecutive years. Adjusted figures excluding a recurring charge are not adjusted, they are wrong.

10. Return on TANGIBLE capital is Munger's first question, and the spread is the story. Corteva earns 31.7% on tangible capital and 4.5% on accounting equity. That gap is not noise — it is the price a previous buyer paid, parked in goodwill and intangibles. Excellent business, bought dearly by someone else. Watch for a near-zero tangible denominator making the ratio meaningless.

11. Capex ≫ D&A means owner earnings are understated, because the excess is growth capex, which belongs to the owner. Capex ≪ D&A on a capital-intensive business means the opposite — it is under-investing and the bill is deferred.

12. Holding is not buying. Buffett's Coca-Cola basis is ~$3.25; he is not buying at $92. "Buffett owns it" is never evidence a price is sensible. State which question is on the table: initiate, add, or hold.

13. A verdict here does not govern a trade with a date on it. This framework has nothing to say about a three-week holding period. Do not let an intrinsic-value number override a pre-committed dated plan — see trade-journal, whose gates were written before conviction was tested. What this framework uniquely supplies is the price of the fallback: "if the trade fails I'll hold it long term" is only available when intrinsic value is near the price. When it is half the price, that exit does not exist and the written stop is the only floor. Say so — it is the most useful sentence this skill produces.

14. A stale valuation gate must be RE-BASED, not obeyed or ignored. A ceiling written as "19.5x forward = $227" is a multiple gate wearing a price tag. When the fiscal year rolls and new guidance lands, recompute it — 19.5x FY2027 guidance of $12.00–12.25 is $234.00–$236.44, so a $232.45 price that looks like a breach is actually inside the gate. Show the arithmetic; never silently drop a gate you wrote.

Output shape

Four filters (stop at the first fail, see buffett-checklist) → owner earnings and how they were built → return on capital → what the market is asking, in owner-earnings terms → reverse DCF → calibration against a Buffett holding → one-line verdict naming initiate / add / hold, with the fallback price stated. Lead with the number; keep the chain loaded for "why".

Related

sec-financials supplies every figure. buffett-checklist owns the qualitative filters, management read and temperament rules — run it alongside, not instead. price-history locates a name in its own cycle. trade-journal holds the written invalidation this must be checked against.