Four traps, four companies. Figures from SEC filings and XBRL data, checked 20 Nov 2026; current prices are 19 Nov 2026 closes.
Trap 1: peak earnings make a low P/E (Micron)
Memory chips go through booms and busts, so Micron is a cyclical business. Its diluted EPS and August share price each fiscal year:
| Fiscal year | Diluted EPS | August close | P/E |
|---|---|---|---|
| 2018 | $11.51 | $52.52 | 4.6 |
| 2019 | $5.51 | $45.27 | 8.2 |
| 2020 | $2.37 | $45.51 | 19.2 |
For 2018: 52.52 ÷ 11.51 = 4.6. The P/E was lowest at the top, and 2018's profit didn't last: earnings fell by four-fifths in two years. By December 2018 the share price was $31.73, down 40% from August.
What to do: use normalised earnings, such as the average over a whole cycle (Damodaran's chapter in Go deeper).
Micron's fiscal 2026 EPS (year ended 3 Sep 2026) is $74.33: a P/E of 9.3 at $690.44. Is that a peak? A P/E can't tell you. You'd look at memory prices, new supply and how far margins sit above their average.
Trap 2: one-off items move earnings (Coca-Cola)
Coca-Cola's net income rose from $10.631B in 2024 to $13.107B in 2025, up 23%. Revenue rose only 1.9%. The notes to its 2025 10-K explain most of the gap:
| One-off item | 2024 | 2025 |
|---|---|---|
| Other operating charges, total | $4.16B | $1.26B |
| of which fairlife payment remeasured | $3.11B | $0.05B |
| Gain on selling its Coke Consolidated stake | — | $1.95B |
| Charge on Africa bottlers held for sale | — | $1.27B |
The fairlife charge: Coca-Cola owed fairlife's former owners a final payment tied to fairlife's results, and each rise in the expected payment was booked as a cost. It paid $6.17B in March 2025. The Africa charge is a write-down: bottlers Coca-Cola has decided to sell ("held for sale") are written down to what it expects to get for them, less selling costs.
So most of the jump is a smaller charge in 2025, not a better business. On 2025 EPS of $3.04, the P/E is 31.6 at $96.09. That is real GAAP profit, but 23% is not a growth rate you can extend. (Gains work the same way: about 11% of Nvidia's TTM pre-tax profit came from gains on shares it owns in other companies.)
Trap 3: negative equity breaks price-to-book (Starbucks)
Starbucks' shareholders' equity was −$7.67B on 27 Sep 2026 (−$8.10B a year earlier).
market cap = 104.47 × 1.1434B diluted shares = $119.5B
P/B = 119.5 ÷ −7.67 = −15.6 → meaningless
It comes mainly from retained earnings of −$8.58B: over the years Starbucks paid out more in buybacks and dividends than it kept in profit, partly with borrowed money. The business isn't failing (diluted EPS was $2.21 in fiscal 2026), but P/B and return on equity stop working. Use earnings or cash-flow multiples instead.
Trap 4: "adjusted" EBITDA leaves out a real cost (AMD)
EBITDA, as in lesson 04, is operating income plus depreciation and amortisation. AMD also publishes its own Adjusted EBITDA. For the first nine months of 2026:
plain EBITDA = operating income 6.298
+ depreciation and amortisation 0.666
+ amortisation of acquired intangibles 1.629
= $8.59B
AMD's Adjusted EBITDA = $10.22B
gap = 1.63B, of which stock-based compensation = 1.51B (93%)
Adjusted EBITDA is 19% higher, so an EV/EBITDA built on it comes out about 16% lower. But paying staff in shares is a real cost (Module 3), and SEC guidance warns that adjustments can make a measure misleading.
None of this makes a stock cheap or expensive. It makes sure the multiple measures what you think.