Additions 2026-09-23 — NLY (Annaly Capital Management), industry stages, 2026-09-23
Cluster pass, 23 Sep 2026.
Additions 2026-09-23 — NLY (Annaly Capital Management), industry stages, 2026-09-23
Slot cluster-additions-2026-09-23, member NLY only, split into its own file as the slot note allows ("Split into per-name cluster files if that is cleaner"). Run by hand in a session on the owner's request ("run annaly capital via pipeline skill"), not by the routine. Model-drafted; nothing here is human-verified. Every figure carries the filing it came from; [background] marks background knowledge and [search-summary] marks anything that came from a search result. HHH and 1913.HK are not scored here and stay pending in the slot.
Stages: S0 (lens), S4, S6, S7. Instrument: SKILL.md §4.3, risk-carrying balance sheets, as the watchlist row directs. How that fits is itself a finding: see §1.
§0 — Sources and how quotes were taken
Rung 1 (SEC EDGAR, User-Agent sent) answered everything filing-based: the FY2025 10-K (accession 0001043219-26-000013, filed 2026-02-12), the FY2023 10-K (0001628280-24-005115, for the 2021–2022 columns), the Q2 2026 10-Q (0001043219-26-000058, filed 2026-07-29), the 2026 DEF 14A (0001104659-26-052954), the 8-Ks of 2025-12-22 (ATM agreements), 2025-12-29 (Chief Legal Officer retirement) and 2026-08-19 (Series I redemption), the 424B5 of 2025-12-22, and XBRL companyfacts. Equibles (rung 4): the Q1 and Q2 2026 call transcripts and FINRA short interest, 3 calls. 3spread (rung 4b): holders, owners, insiders, sales. ROIC.ai (rung 5): prices. Two WebSearch queries (rung 6) for scuttlebutt.
Each filing was converted to text with whitespace collapsed, and every quoted span below was located in that text by exact substring match before it was written. Transcript content is paraphrased and never put in quotation marks: the Equibles text is machine speech-to-text with visible transcription errors ("Annalise", "hook value"), and a quote from it would be a quote of the transcriber.
§1 — The finding: the instrument fits a spread book on two tests out of four, and the two that fit are the two it fails
§4.3 was written for underwriters. Annaly carries risk on its own balance sheet — rate, basis, prepayment and, in its residential credit and MSR books, credit — but it has no reserves and no premium. Two of the four tests have a spread branch that fits directly (test 2, test 4); two need a translation (test 1, test 3). Scored as translated below, with the translation stated each time, so the next variant can be written from it rather than improvised again.
The number that matters, from the company's own tables: between the end of 2021 and the end of 2025 Annaly reported $12.71 per share of earnings available for distribution (EAD) — its headline non-GAAP earnings, the figure the dividend is set against — and paid $11.52 of common dividends. Book value per share went $31.88 → $20.21. Add the dividends back and a holder's economic result over those four years was −$0.15 per share, roughly zero, on a $31.88 start.
| Year | EAD / share | Dividend declared / share | BVPS year-end | Economic return (computed) |
|---|---|---|---|---|
| 2021 | 4.65 | 3.52 | 31.88 | — |
| 2022 | 4.23 | 3.52 | 20.79 | −23.7% |
| 2023 | 2.86 | 2.60 | 19.44 | +6.0% |
| 2024 | 2.70 | 2.60 | 19.15 | +11.9% |
| 2025 | 2.92 | 2.80 | 20.21 | +20.2% (company reports 20.2%) |
| H1 2026 | 1.55 | 1.45 | 20.15 | company reports 6.9% |
EAD and BVPS: FY2023 10-K (2021–2023 columns) and FY2025 10-K (2024–2025), Net Income (Loss) Summary; Q2 2026 10-Q, same table. Dividends: XBRL CommonStockDividendsPerShareDeclared, FY 10-K facts; H1 2026 = $0.70 + $0.75 (Q1 call, Q2 10-Q MD&A). Economic return = (ΔBVPS + dividends) / opening BVPS, computed here except where the company's figure is given.
So the EAD line is not a measure of what the business earned for its owners across a rate cycle; it is a measure of carry. The difference went through book value. That is what test 1 asks about, and it scores 0.
And the book grew by issuing shares. Diluted weighted-average shares (XBRL, split-adjusted by the company for the 2021 1-for-4 reverse split): 353.7m (2020) → 641.0m (2025) → 740.3m (Q2 2026); 753.7m outstanding on 2026-07-24 (10-Q cover). Common stock issued for cash, per the cash flow statement: $7.78bn over 2022–2025, plus $955.7m in H1 2026 through the ATM. Common equity (total stockholders' equity less $1.54bn / $1.80bn of preferred) went $11.63bn (2021) → $15.12bn (Q2 2026), +$3.49bn — so roughly $5.2bn of the $8.7bn raised was absorbed by the book rather than added to it. Test 4 scores 0.
What the pass also has to say, as plainly: since 2023 the picture has turned. Economic return +6.0%, +11.9%, +20.2%, then +6.9% in H1 2026; EAD above the dividend for nine consecutive quarters (Q2 2026 10-Q); and the recent issuance is above book — H1 2026 ATM proceeds of $955.7m for 42.6m shares are $22.43 a share against a $20.15 book, accretive on its face. The score is a five-year score. The two 0s are about what the model did through the last hiking cycle, which is the only cycle in the window.
§2 — NLY · mortgage_reit · ✅ lens applied — 1/4
Instrument: §4.3 risk-carrying balance sheets, spread branches where they exist.
| # | Test | Score | Evidence |
|---|---|---|---|
| 1 | Reserve honesty — are the reported earnings real? | 0 | Translation: an mREIT has no loss reserve; its "reported earnings" are EAD, and the honest check on them is the economic return, which marks the book. 2022–2025: EAD $12.71, dividends $11.52, BVPS −$11.67 → economic result −$0.15/share (§1 table). The spread-writer analogue of "adverse development in two of five" is a year in which EAD covered the dividend while the book lost a fifth of its value: 2022, EAD $4.23 against a $3.52 dividend and a −23.7% economic return |
| 2 | An edge that survives the cycle (spread branch) | 0 | Net interest spread (excluding PAA, i.e. including swap carry) 1.89% / 1.70% / 1.32% / 1.22% / 1.40% (2021–2025, FY2023 and FY2025 10-Ks) — positive in 5 of 5 but not stable: it compressed 35% into 2024 and the dividend was cut 26% ($3.52 → $2.60) in 2023. The GAAP spread, without swap carry, was −0.81% (2023) and −0.23% (2024). Numbers removed to get from GAAP to the economic figure: the net interest component of swaps, $1.59bn in 2023 and $1.25bn in 2024 (GAAP interest expense less economic interest expense, FY2025 10-K). And the swap carry is fading: the CFO named lower swap income as the offset to EAD in both Q1 and Q2 2026 (calls, paraphrased) |
| 3 | Capital that is theirs, and rated | 1 | Translation: premium-to-surplus becomes leverage; ceded share becomes funding that is not the company's. Economic leverage 5.7x / 5.5x / 5.6x (2023–2025) and 5.6x at Q2 2026, inside a policy the 10-K states as: "We generally expect to maintain an economic leverage ratio of no greater than 10:1 considerate of our overall capital allocation framework." (FY2025 10-K, Item 1, Leverage). $9.6bn of assets available for financing, ~57% of capital (Q2 call, CFO, paraphrased). Funding spread across 37 secured-financing counterparties — 23 North America, 10 Europe, 4 Japan — on $88.0bn (Q2 2026 10-Q, Counterparty Risk Management). The 1-in-250 analogue: the company's own +25bp MBS spread shock is −8.7% of NAV (Q2 2026 10-Q, Item 3 table). Scored 1 on those numbers, with the structural caveat stated in the falsifier: the funding is repo with a 33-day weighted average maturity (Q2 call) — rented by construction, every month |
| 4 | Growth that does not dilute | 0 | Diluted shares 353.7m → 740.3m (2020 → Q2 2026, 2.09x) while BVPS fell $31.88 → $20.15 (2021 → Q2 2026). $8.7bn of common raised 2022–H1 2026 against +$3.49bn of common equity (§1). The recent raises are above book and management calls them accretive ($280m of accretion, Q2 call, CEO, paraphrased) — that is a 2024–2026 fact, the test is a five-year one |
Cycle question — where is the line in its cycle, on a named measure, and does the case need it to stay? The measure is the mortgage basis (current-coupon MBS over the rate curve), which Annaly publishes: 137bp (Dec 2023) → 126bp (Dec 2024) → 87bp (Dec 2025) → 106bp (Mar 2026) → 94bp (Jun 2026) (FY2025 10-K and Q2 2026 10-Q, interest rates and spreads tables). It is near its tightest in the three-year window, held there by a policy buyer: the GSEs' $200bn MBS purchase mandate announced in January 2026, of which roughly $45bn had settled through May (Q2 call, CEO, paraphrased; the $200bn figure also in [search-summary]). Yes, the case needs the basis not to widen. At the company's own sensitivity, a return to the Dec 2024 basis (+32bp from June) is roughly −11% of book before hedging adjustments (linear extension of the −8.7% per +25bp row — an extrapolation, the table stops at 25bp). A name that works only while a policy buyer holds the spread tight is a cycle trade, and S7 and S8 are told so.
The AI line (one sentence, not scored). Cheap intelligence does not set the MBS basis; it lowers the cost of the underwriting, servicing oversight and correspondent onboarding that Annaly buys from partners, which helps an "operating light" capital provider more than an origination-heavy one.
Falsifier — one number. Book value per share below $18.45 — the Q2 2025 figure (Q2 2026 10-Q), the lowest quarter-end BVPS in the filings read. Getting there from $20.15 is −8.4%, one +25bp basis shock at the company's own sensitivity.
Liquidity. ADV $183.1m/day (21 sessions 2026-08-24 → 2026-09-22, close × volume, ROIC NYSE:NLY). Last price $21.44 (2026-09-22). P/B 1.06 on the 2026-06-30 BVPS of $20.15; market cap ~$16.2bn on 753.7m shares. Dividend $0.75/quarter, 14.0% annualised on the last price.
Re-tag check (§4.1). mortgage_reit holds one member, so there is no mis-clustering to fix. If the owner adds AGNC, RITM, STWD or a peer, test 1's translation and test 3's are the ones to carry; Annaly's own 50/30/20 target (Agency / residential credit / MSR, Q1 call, paraphrased) means peers will differ mostly on test 3.
§3 — S4 Industry & supply map
The chain, in words. Borrower → originator (now mostly non-banks; independent mortgage banker net margins were 21bp in 2025, Q1 call, Mike Fania, paraphrased) → guarantor (Fannie Mae, Freddie Mac, Ginnie Mae) → Agency MBS → holders (money managers, banks buying CMO floaters, overseas buyers ~$65bn YTD, the GSEs' own retained portfolios under the 2026 mandate, mREITs) → funded by repo dealers (37 counterparties for Annaly) → hedged with swaps and Treasury futures (the CEO described a hedge book of roughly two-thirds swaps to one-third Treasuries, Q1 call, paraphrased). Beside it, two chains where Annaly is not a price-taker: residential credit, where Onslow Bay buys non-QM/DSCR loans from ~350 correspondents and securitises them on the OBX shelf (13 deals / $6.8bn UPB in Q2 2026, 10-Q MD&A); and MSR, bought in bulk from originators and serviced by third-party subservicers at a per-loan fixed cost.
Where the pricing power is. In Agency, nowhere on Annaly's side: it is a price-taker in the deepest fixed income market after Treasuries, and the price-setter in 2026 is the US government (the GSE purchase mandate) plus the Fed's balance sheet policy. In residential credit there is some: scale on the shelf gives lower fixed and underwriting costs per deal and tighter pricing than most issuers (Q2 call, Fania, paraphrased — a company claim). In MSR it is structural and specific: buying servicing released from sellers who would otherwise be left with stranded servicing cost, without competing for units on its own platform (Q2 call, Ken Adler, paraphrased — a company claim).
Concentration. No customer concentration in the usual sense. The counterparty concentration is the funding: $88.0bn of secured financing across 37 counterparties, $71.2bn of it in North America (Q2 2026 10-Q). The policy concentration is the guarantee: the 10-K lists as a risk factor that "Any new laws, regulations or administrative actions modifying the relationship between Fannie Mae, Freddie Mac and the federal government could affect our business model or business operations." (FY2025 10-K, Item 1A, Compliance, Regulatory & Legal Risks).
Who wins share, from whom. Non-banks from banks in origination (secular, per both calls); private-label securitisation from the GSEs at the margin — non-Agency gross issuance >$150bn YTD, up ~50% (Q2 2026 10-Q MD&A), on pace for its largest year since 2007. Annaly is positioned on the winning side of that second shift. Banks may take back some mortgage balance-sheet share under the 2026 capital re-proposal (residential RWAs estimated −30%, Q1 call, CEO, paraphrased), which is a positive for Agency spreads and a competitor for MSR.
§4 — S6 Scuttlebutt
Outside evidence 1 — the board's own claim row. Chris Whalen on The Julia La Roche Show, 2026-09-19 (pipeline/gold/candidates/julia-la-roche-2026-09-19-410-chris-whalen-age-of-uncertainty-falling-05-nly.yaml, bullish, conviction 4, machine-extracted): Annaly is one of his bigger positions, he owns the common not the preferreds, his basis is about 0.9x book, and he expects it to be a survivor of a coming mortgage consolidation because it is an investor rather than an originator. His stated condition is that they maintain the dividend. Two things to take from it: a buyer with a 0.9x-book basis is a useful reference point against today's 1.06x, and his thesis is the company's own "operating light" line, not an independent one.
Outside evidence 2 — sell-side, [search-summary]. BTIG upgraded NLY to Buy (and cut AGNC to Neutral) in January 2026 as MBS spreads reached multi-year tights — a relative call on the diversified model over the pure Agency one (Seeking Alpha headline via WebSearch). A Motley Fool piece of 2026-05-20 is titled "Annaly Capital's Book Value Tells the Story Its Dividend Yield Doesn't" (title only; the article was not read).
Short interest. 20.66m shares at the 2026-08-31 settlement, 2.7% of the 753.7m shares outstanding, 3.8 days to cover (FINRA via Equibles). Not a crowded short.
No short report found in two searches; a failed search is not evidence there is none.
Product test / expert call: ⏳ OPEN (user). For an mREIT the useful expert call is a repo desk or an MBS strategist on how long the GSE bid lasts.
§5 — S7 Thesis and framework test
Framework used: a plain cycle read (the lens table from _clusters/ai-capex-2026-09-20.md is for AI/semis and does not apply). Bear case first.
Bear case. The dividend has been paid, in the only full rate cycle in the window, out of carry while book value absorbed the losses — 2022–2025 EAD of $12.71 against an economic result of roughly zero. The model then refilled the book by issuing shares (2.09x the diluted count since 2020). Today the basis is at 94bp, near its three-year tight, held there by a government buyer with a finite mandate; the swap carry that kept the economic spread positive in 2023–24 is rolling off (named by the CFO in both 2026 calls); and the stock trades at a premium to book, which is what keeps the ATM machine accretive. If the basis widens, book falls, the premium closes, the ATM turns dilutive, and the loop that has made 2024–26 look good runs in reverse. Insiders have been net sellers at $22.5–23.1 through 2026 (S5 in the scorecard).
Bull case. The business changed shape. Residential credit and MSR are now ~43% of capital (22% + 21%, Q2 2026 10-Q) against a 50/30/20 target, and neither is a pure basis trade: OBX is a securitisation franchise with scale advantages the company can show (largest issuer of expanded-credit mortgages per the Q2 10-Q; a $900m+ average non-QM deal per the Q2 call), and a 3.3% note-rate MSR book is a natural hedge against the Agency book's prepayment and rate exposure. Leverage is 5.6x, low for the sector. Nine straight quarters of EAD above the dividend, a dividend raised in Q2 2026, and issuance above book.
Variant perception. The market prices NLY as a yield (14.0%) and a book (1.06x). The variant is that the right unit is neither: it is economic return per share over a full cycle, which was about zero for 2022–2025 and is running at a double-digit annualised rate now. The question the price does not ask is how much of the 2024–26 recovery is the new mix (resi credit, MSR) and how much is the basis tightening from 126bp to 94bp under a policy bid. If the mix is doing the work, the book holds when the GSE bid ends; if the basis is, it does not.
Kill criteria (specific, testable):
- BVPS below $18.45 at any quarter-end (the Q2 2025 figure, the lowest quarter-end BVPS in the filings read).
- EAD below the dividend in two consecutive quarters. The streak is nine; management itself said there may be quarters slightly below it (Q2 call, paraphrased).
- Economic leverage above 7.0x without a stated reason. It has sat at 5.5–5.8x since 2023.
- ATM issuance while P/B is below 1.0. Management's own first criterion for raising is price-to-book being accretive (Q2 call, CEO, paraphrased). A raise below book would say the criterion gave way.
Time horizon. 2–4 years, one GSE-mandate cycle: the $200bn mandate is the thing to watch run off.
Conviction: ⏳ OPEN (user).
§6 — What this pass could not do
- The Q3 2026 book value is not known. Price fell from $23.29 (2026-08-20) to $21.44 (2026-09-22), −7.9%, on no NLY filing other than the Series I redemption notice; the cause was not established here and is not inferred.
- No investor supplement or investor-day deck was read (gate-minimum). The per-segment returns quoted by management (Agency 14–16%, resi upwards of 15%, MSR upwards of 13%) are from the Q2 call and are not reconciled to a filing.
- §4.3 fit. Tests 1 and 3 were scored under translations written in this file. The owner may want a fifth variant for spread books (mREITs, and the life/annuity spread writers already scored 0/4 and 0/3 in
cluster-insurance-rescore-2026-09-22.md) rather than continuing to translate.