The full $281.4B book, repriced slice by slice: 140 verified named slices at their disclosed par and internal rating, plus the undisclosed remainder as sector × rating-band cells. No scaling from the CUSIP tape, no single opaque multiplier. Guaranty claims are non-accelerated — losses emerge as scheduled debt service comes due, so every scenario converts ultimate loss to present value over each slice's remaining life before it touches equity.
S&P definitions: AA severe = GDP −15%, unemployment 20% · AAA extreme = Great Depression rerun.
Severity scales the idiosyncratic overlays on named stressed credits. Discount prices the non-accelerated claim stream.
Adjusted book value absorbs the after-tax present value of incremental stressed losses — incremental over the $321M already reserved.
The $281.4B book tiled exactly: verified named slices plus residual sector × band cells. Rates are % of slice par.
Base rates are calibrated so the book's base-case loss equals AGO's disclosed $192M expected loss. Scenario rates = base × multiplier, capped at sector LGD. Multipliers are judgment — the table is the model.
Named exposures driving the stressed loss. verified = par and rating re-extracted from the filed 2Q26 supplement. overlay = a judgment-based idiosyncratic loss rate floors the band rate. judgment = analyst assumption, not disclosure.
1. Stratified book, no scaler. Every dollar of the $281.4B book sits in exactly one modeled slice: 140 named slices (138 credits; Palomar Health and University of Essex split into BIG slice + remainder so nothing is double-counted) at verified net par and AGO internal rating, plus 13 residual sector × rating-band cells computed as the book's published stratification minus named par. The CUSIP tape ($38.66B original issue par) is documentation of indenture coverage, not a loss input — it is never scaled and never conflated with net par.
2. Calibration, not assumption. The base case reproduces AGO's own booked number: per-band base loss rates are set so the book's base ultimate loss equals the disclosed $192M net expected loss to be paid (2Q26). Only the distribution across bands uses judgment (relative risk weights: AAA 0.05, AA 0.15, A 0.5, BBB 1.5, BB 8, B 25, CCC 60). The calibration is shown in the band table — it is not tuned to produce any particular share price.
3. Stated scenario multipliers. Stress scenarios multiply base rates per band (see band table), capped at sector LGD × 0.98. LGDs are judgment: US public finance 35%, non-US 45%, structured 60%. The multipliers are the model's explicit stress view — change the scenario, read the table, see exactly what moved.
4. Idiosyncratic overlays (judgment). Five named stressed credits carry judgment ultimate-loss rates that floor the band rate, scaled by the severity dial: Brightline 30% (2Q26 transcript: liquidity pressure, proactive creditor talks), PREPA 30% (restructuring unresolved), Thames Water 15% (special-administration tail risk), Palomar BIG slice 25% (forbearance), Westchester Medical 12%. AGO discloses expected loss only at book level — every name-level allocation here is analyst judgment, labeled as such.
5. Non-acceleration. Guaranty policies pay scheduled debt service, not lump sums. Ultimate loss converts to present value as a level annuity over each sector's assumed remaining life — US public finance 14y, non-US 12y, structured 7y (judgment) — discounted at the claims discount rate: PV = L × [1 − (1+d)^-T] / (d·T). This is timing, not an instant par hit.
6. Share price. Stressed ABV = ABV − (book PV loss − already-reserved $321M) × (1 − tax). Divided by shares outstanding. The $321M is the PV of net expected loss to be expensed (2Q26); tax is stylized at 21%.
Presets. Base = AGO's booked expectation (severity 0%). Great Recession replay: unemployment 10%, GDP −4%, home prices −30%. S&P AA (severe) and AAA (extreme) per S&P Global Ratings Definitions (Oct 2024): AA envisions GDP declines up to 15%, unemployment up to 20%, equities −70%; AAA is a Great Depression rerun — real GDP −26.5%, unemployment peaking at 24.9%. S&P affirmed AGO at AA in July 2026 noting capital adequacy redundancy above the AAA stress level — this lab lets you check that claim yourself.
Named coverage is ~34%, not 100%. 138 named credits ($94.7B) carry verified par and rating from the filed 2Q26 supplement. The remaining $186.7B is modeled as undisclosed granular residual by sector and rating band — stratification, not omniscience. Private/144A names (JFK New Terminal One, Brightline) and non-US exposures have no public indenture by structure; they are disclosure records, not tape gaps.
The tape is original-issue par. The 2,413-CUSIP tape totals $38.66B of original issue par. It is not current AGO net par, it is not a loss input here, and the two are never added or compared.
Name-level losses are judgment. AGO discloses expected loss only at book level ($192M to be paid; $321M PV to be expensed). Every allocation to a named credit — band rates, multipliers, overlays — is analyst judgment. The model cannot see AGO's non-public surveillance, reinsurance recoveries in detail, or future refundings.
Scenario analysis, not a forecast. Not investment advice. Multipliers, LGDs, remaining-life and overlay assumptions are stated above so you can disagree with them precisely.
The AAA (extreme) preset is S&P's Great Depression rerun: real GDP −26.5%, unemployment peaking at 24.9%. A municipal default wave on that scale does not happen inside a functioning credit system. It happens in a world where no commercial bank survives in its current form, money markets freeze, and the Treasury and Federal Reserve intervene at a scale beyond 2008. In that world, AGO's modeled $9.05B ultimate loss would be a symptom of a systemic collapse, not the story. The honest question is not whether the equity survives the loss, but what the guarantor's position looks like in the world that produces it.
A guarantor is not a bank. A bank dies from a funding run; AGO cannot have one — no deposits, no repo, no commercial paper. Policies pay scheduled debt service as it comes due (non-accelerated), so a $9B ultimate loss arrives as a stream spread over decades, funded from claims-paying resources and investment income, not as a single draw. After paying a claim, AGO subrogates to the bondholders' position and recovers in the restructuring. And stress is historically when the franchise strengthens: AGO survived 2008 with its ratings and wrote new business at crisis-era pricing in 2009–2011. The same credit freeze that breaks a bank's balance sheet reprices the guarantor's product upward.
The mitigation playbook is precedented, not hypothetical — it is what the company has done before:
In extremis, AGO can buy reinsurance or stop writing business and run off the book while returning capital — the "shut the lights out" option. Mitigation compresses losses; it does not erase them. At AAA-extreme, claims-paying resources cover stressed ultimate loss only 1.1× — solvent on paper, with no margin if the path is worse than the endpoint. And the market prices the path, not the endpoint: the equity would trade at a deep discount to stressed ABV long before $9B of losses actually materialized. The table shows the math at the end of the stress; it does not show the mark-to-market violence of getting there.