Playbook › Playbook
Dividend Yield Theory inverts into a buy signal exactly when the business is breaking
Claim
analysis_notes.md §3 lists Dividend Yield Theory as a model to weight for dividend payers:
buy when the yield is high versus its own history. The arithmetic is yield = dividend ÷
price, so a yield can double two ways — the numerator doubled (the business is
compounding) or the denominator halved (the business is breaking). DYT reads both as
identical, and the second one is where it gets you killed.
Worse, the signal strengthens as the situation deteriorates. Every incremental drop in the price widens the gap to the historical band and makes DYT shout louder.
Evidence — NVO, 2026-07-31
| Input | Value |
|---|---|
| TTM dividend/ADR | $1.859 |
| Price | $47.08 |
| Current yield | 3.95% |
| 5-year average yield | 1.79% |
| Ratio | 2.21x |
Naive DYT fair value ($1.859 ÷ 0.0179) |
$103.85 |
DYT said a $47 stock was worth $104 — a 121% upside call — on the day after a Phase 3 failure, in the middle of a −66% drawdown, on a company guiding operating profit down 4–12%. The entire signal came from the denominator: the dividend rose 15%, the price fell 66%.
Three things made it a trap rather than an opportunity, and all three are checkable in advance:
- The payout policy was not progressive. Novo pays ~50% of net income by explicit Danish policy. When earnings fall, the dividend falls. DYT's mean-reversion assumption requires a dividend that holds while the price recovers; here the dividend follows the earnings down and the yield "normalises" by falling, not by the price rising.
- The historical band was itself an artifact of a boom. The 1.79% five-year average was set during the 2021–24 GLP-1 melt-up when the stock traded 24–40x earnings. It was never a fair-value yield — it was a bubble yield. Reverting to it means re-inflating the bubble.
- FCF did not cover the dividend. FY2025 dividends were 179% of free cash flow.
Guard
Before running DYT, answer three questions. If any answer is wrong, do not report a DYT number at all — or report it and explicitly discard it, so the reasoning is on the record.
| Question | Disqualifying answer |
|---|---|
| Why did the yield rise? Decompose it: how much from dividend growth, how much from price decline? | Mostly price |
| Is the payout policy progressive or formulaic? | A fixed payout ratio — the dividend will fall with earnings |
| Was the historical band set during a boom? | Yes — the band is a bubble artifact, not a fair-value anchor |
| Does FCF cover the dividend? | No |
The repair, when DYT is disqualified: don't discard the yield lens entirely — reset the band to the peer class the company is becoming, not the one it used to be in. On NVO, applying a mature-pharma yield of 3.5–4.5% (Pfizer/Merck/BMY territory) to the same $1.859 dividend gives $41–53 — a bracket that contained the actual price and agreed with the earnings-power anchor. The naive band said $104; the reset band said $47. The reset version is the one worth reporting.
Refinement — the guard can SPLIT, and the window decides question (a)
VICI, 2026-08-03. The first case where the guard did not return a clean verdict. Two questions passed, two failed — and question (a) returned opposite answers depending on the measurement window, with both answers true.
| Window | Yield move | Dividend share | Price share |
|---|---|---|---|
| Trailing 12 months (from the $34.01 high, rate $1.732) | 5.09% → 6.85% | 13.0% | 87.0% |
| Full 5-year band period (Aug 2021: $30.91, rate $1.32) | 4.27% → 6.85% | 65.6% | 34.4% |
Over twelve months VICI looks like NVO: a price-driven yield, disqualifying. Over the band period it looks like a compounder: the dividend grew 36% while the price fell 15%. Both computations are correct. They answer different questions, and only one of them is the question DYT actually asks.
The rule this produces: decompose over both windows — the window the historical band was measured over, and the recent window. Then read them as two separate facts:
- The band-period decomposition tells you whether the yield LEVEL is earned. If the dividend did most of the work over the band period, the high yield is a real dividend achievement, not a broken price.
- The recent decomposition tells you what the market is currently saying. A price-driven recent move is a warning about the forward thesis regardless of how good the history was.
They can disagree, and when they do the correct response is not to pick one — it is to report both and let questions (b) and (d) break the tie. On VICI they did: the payout was progressive in practice (7 straight raises, a 73–78% payout band rather than a fixed formula, AFFO still growing) and cash coverage was 1.37x on AFFO. The pitfall note's own stated distinguishing test — "defended by policy and covered by cash, not whether the yield looks high" — is what resolved it. That test is doing more work than the four-question table implies and should be treated as the primary filter, with the table as the checklist.
Question (c) still disqualified the naive number on its own, for a reason worth naming separately from the NVO boom-artifact case: VICI's 5.15% five-year average band spans two different rate regimes (a 1.30% 10yr in 2021, a 4.68% 10yr today). Reverting to it would demand a 47bp spread over Treasuries for a BBB− REIT with 70% tenant concentration. A band does not have to be set in a bubble to be invalid — it only has to be set in a different rate regime. For any rate-sensitive income asset, add that as a distinct failure mode of (c).
The prescribed repair worked exactly as written. Two independent resets converged:
| Reset | Implied price |
|---|---|
| Naive band (5.15%) | $34.95 ❌ |
| Spread-to-Treasury at the 182bp period average | $27.69 |
| Peer class it is becoming (gaming net-lease, 6.5–7.4%) | $24.32 – $27.69 |
The reset moved the answer $7.26 per share (−21%), against a weighted fair value of $27.75 and a market price of $26.27. The reset version was right; the naive version was the largest single error available in the file.
Why this is the same failure as the drawdown fallacy
[[principle-down-a-lot-is-not-cheap]] says: measure against the name's own multiple band, not its 52-week high, because a falling price flatters a drawdown statistic. This is the identical error one layer down — a falling price flatters a yield statistic too. Any metric with price in the denominator improves automatically as the thesis breaks. Treat every one of them as a question, not an answer.
What would falsify this
Nothing about the arithmetic. What would refine it: a case where DYT fired on a price-driven yield expansion and was right — i.e. the price genuinely overshot while the dividend held. Those exist (cyclical dividend payers at trough sentiment), which is why the guard is a decomposition rather than a blanket ban. The distinguishing test is whether the dividend is defended by policy and covered by cash, not whether the yield looks high.
Related
[[principle-down-a-lot-is-not-cheap]] — the same denominator fallacy on multiples. [[principle-story-vs-revenue]] · [[principle-primary-source-beats-vendor]].
History
- 2026-07-31 — found in the NVO analysis. DYT returned $103.85 against a $47.08 price and a $36–50 weighted fair-value range; it was the single most wrong model in the file and would have flipped the verdict from HOLD to BUY had it been weighted mechanically.