ALLThe Allstate Corporation

ALL — dossier, 2026-09-23

Slot names-insurance-2, first member. Model-drafted unattended by the cloud research routine. Nothing here is human-verified. Every figure carries the document it was read from; anything from background knowledge is marked [background]. Quotes were located by exact substring search in a text extraction of the filing and copied out — never re-typed.

Written after ACGL, KNSL and AX (slot names-insurance-1, 2026-09-22) and after the Stage 0 re-score (_clusters/cluster-insurance-rescore-2026-09-22.md, which scores this name 3/4).

---

1. What is new since the last filing anyone has read

The Q2 2026 10-Q was filed 2026-08-05. Two Reg FD 8-Ks have landed since, and they contain the most decision-relevant facts on this name:

Against the same two months a year earlier, from the equivalent releases:

Two months of Q3 2026 are 2.6x the entire prior-year quarter, with September unreported. The September release is due around 2026-10-15 on the observed cadence (the monthly 8-K has landed on the 15th–21st of the following month in each of the last fourteen months).

Market reaction, stated as correlation and not as cause. ALL closed $252.21 on 2026-09-17, the day the August release was filed, and $229.50 on 2026-09-22 — −9.0% over three sessions, with 2026-09-22 trading 5.26m shares against a 21-session average of about 1.7m and a $227.25–$242.16 range (ROIC NYSE:ALL daily closes). Other things happened in those sessions; this dossier does not claim the release caused the move, only that the move followed it on volume.

---

2. Why the reported margin improved, and how much of it is repeatable

From the Q2 2026 10-Q, Property-Liability GAAP operating ratios (MD&A, ratio table):

The reported improvement is 9.9 points over the half. The underlying improvement is 1.7. The rest is weather and reserve releases — and the 10-Q states plainly:

"Prior year reserve reestimates, including catastrophes, decreased reserves by $641 million in the second quarter of 2026 and $1.65 billion in the first six months of 2026." — Q2 2026 10-Q, MD&A, paragraph following the Property-Liability underwriting results table

H1 2025 was $621m. The release roughly tripled.

For the catastrophe side, Allstate supplies its own benchmark in the same filing:

"The ten-year average effect of total catastrophe losses on the total combined ratio was 13.5 points and 11.0 points in the second quarter and first six months of 2026, respectively." — Q2 2026 10-Q, MD&A, footnote (1) to the combined-ratio table

H1 2026 ran at 10.0 points against an 11.0-point ten-year average — i.e. below normal, before a third quarter that is running far above it.

The FY2025 series, from the Stage 0 re-score file (_clusters/cluster-insurance-rescore-2026-09-22.md §4, sourced there to Allstate's own ratio tables), puts the ex-cat ex-development combined ratio at 87.3 / 95.6 / 91.7 / 85.6 / 79.7 for 2021–2025. The H1 2026 figure of 79.9 continues 2025's level; the machine is working. The question this dossier raises is not whether it works, but what is being reported on top of it.

---

3. The pay metric, which is where the two halves meet

The annual incentive pool is 70% business-unit profitable growth and 30% "Performance Net Income" (DEF 14A 2026-04-10, CD&A, 2025 plan design). The definition of that measure, on p.83 of the same document:

"Performance Net Income is adjusted to replace actual catastrophe losses with plan catastrophe losses." — DEF 14A filed 2026-04-10, "Performance Net Income" definition, p.83

The list of adjustments on that page removes restructuring, run-off underwriting, disposals, unplanned alternative capital, buybacks differing from plan, Health & Benefits, and anything non-recurring above $20m after tax. Prior-year reserve reestimates are not on the list. For 2025 the metric printed $8,634m against threshold $4,540m / target $5,340m / maximum $6,140m, funding the 30% component at 200.0% (CD&A, 2025 annual incentive table).

So the measure that sets the bonus is immunised against the catastrophe losses now running at 2.6x last year, and fully exposed to the reserve releases that supplied 5.6 points of the H1 combined ratio. That is not an accusation of anything; it is the incentive geometry, read out of the filing, and it is the concrete content of the 0 that §4.3 test 1 gives this name.

---

4. Management on the reserves, in their own words

From the Q2 FY2026 earnings call (2026-08-06, Equibles speaker-labelled transcript), the final analyst question. Andrew Gliederman (TD Cowen) asks about auto prior-year reestimates — "6.6 points this quarter, each of the last four prior quarters you've had sizable releases" — and what is generating them. The answer:

"every quarter we think we get we've got reserves as accurate as we can get them so we don't assume there'll be any more coming ever" — Tom Wilson, Q2 FY2026 call, answer to Andrew Gliederman (TD Cowen)

and the reason offered for the releases is external: tort reform in Florida, Georgia and Louisiana, and New York legislation, making earlier loss-cost estimates too high.

The CFO turn that follows attributes the pattern to "a pretty volatile inflationary period", used-car prices "going up in upwards of 60 percent", and says the process is unchanged and externally audited.

Transcript reliability note, recorded so a later reader does not chase it. This call's speaker labels are not trustworthy: two different names are both tagged CFO, one analyst turn has the next analyst's name on it, one answer is merged into the question turn, and the CEO addresses a colleague as "John" who is not either of the two CFO-tagged names. The quote above is safe because the speaker identifies himself inside the turn ("i'll start and then john can jump in"), but no attribution from this transcript should be carried without that kind of internal check. Nothing else from this call is quoted anywhere in this repo.

Two other things worth having from the same call, not quoted:

• The CEO's closing line is an explicit valuation argument — that at this price you cannot get this combination of operational excellence, sustainable growth and capital generation anywhere else.
• Asked by Rob Cox (Goldman Sachs) whether it is prudent to slow the buyback and hold cash, the answer was a commitment to complete the full $4bn, citing $9.5bn of deployable holding-company capital (the $9.5bn figure is the analyst's, in his question, not management's).

---

5. The exposure base is growing, which the cluster file assumed it was not

_clusters/cluster-insurance-2026-09-22.md §5 wrote the variant perception as "a peak-earnings insurer in a line whose exposure base is structurally shrinking", with a kill criterion built on personal-auto earned premium declining. The company's own last monthly unit table says otherwise:

— 8-K filed 2026-06-18, Ex-99, "Allstate Protection Policies in Force" table.

Premiums agree: Allstate Protection premiums earned $29,720m in H1 2026 against $28,373m (+4.7%), written $30,056m against $29,344m (+2.4%). The first half of the cluster's sentence survives this pass and the second half does not. The scorecard's kill criteria are rewritten accordingly.

The series itself has ended. The May 2026 release is the last monthly one carrying the table:

"As previously communicated, policies in force will be reported in our quarterly earnings release going forward." — 8-K filed 2026-06-18, Ex-99, sentence following the PIF table

This is disclosed, not silent, and it was put to management on the Q2 call by Josh Anker (Bank of America), who opened with "I know that monthly PIF reporting is going away" and used June as the last month he could compute from. The CEO's reply thanked him for "debunking the conspiracy theory that because we stopped doing monthly things, we had a problem." Recorded here as a monitoring cadence change — the unit series drops from monthly to quarterly — and not as a red flag.

---

6. Kill test (S2) — the numbers

FY figures from XBRL company-concept series on the FY2025 10-K (accession 0000899051-26-000031); H1 2026 from the Q2 10-Q.

Diluted EPS H1 2026 $21.73 against $9.85. BVPS $123.38 at 2026-06-30 against $82.40 a year earlier (+49.7%) and $108.45 at 2025-12-31. Trailing-twelve-month return on average common equity 49.1% (10-Q MD&A). Debt to equity 22.2%, down from 24.5%.

Survives on every axis by a wide margin. The kill test is not where this name is decided; the reserve line is.

Buyback: "During the first six months of 2026, we repurchased 8 million common shares, or 3.0% of total common shares outstanding at December 31, 2025, for $1.66 billion." — Q2 2026 10-Q, Capital Resources. The authorisation is $4.00bn from 2026-02-04, with $2.60bn remaining at 2026-06-30 and a 2028-02-29 deadline. At $229.50 that is 1.86x book.

---

7. Ownership (S5)

3spread, 13F report date 2026-06-30 (threespread.py holders ALL; 1,222 filers against 1,668 in the prior quarter — the quarter is incompletely ingested and no exit may be read from an absence; the named non-filers include both Vanguard entities, Franklin, Geode, GQG, Morgan Stanley, Northern Trust, Dimensional, UBS AM and First Trust).

The Sixth Street row is not a holding. 3spread reports Sixth Street Partners Management Company as a NEW holder of 1,856,753 shares, byte-identical to Charles Schwab Investment Management's position, and threespread.py holders now flags it automatically. This is the defect documented in ledger/research/README.md: Sixth Street's 13F-HR accession 0001752724-26-000051 carries Schwab's information table. ALL is the fifth issuer where it has appeared, after QCOM, ACGL, KNSL and AX.

Vanguard. The 13G/A cause found at APH on 2026-09-21 is confirmed here from this issuer's own proxy: "On March 26, 2026, The Vanguard Group, Inc. filed a Schedule 13G/A reporting that on January 12, 2026, it underwent an internal realignment such that The Vanguard Group, Inc. and its subsidiaries or business divisions of subsidiaries will report beneficial ownership separately" — DEF 14A, 5% holders table, footnote (1). Two issuers now carry the same footnote, so it is a firm-wide reporting change and every "Vanguard exited" reading in this repo's 13F data is wrong.

Insiders: 21 directors and executive officers hold 1.55% of the common, and "no director or executive officer other than Mr. Wilson, beneficially owned 1% or more of the outstanding common stock" (DEF 14A footnote 5, as of 2026-03-01). No shares pledged. Form 4 traffic: 172 filings in the twelve months to 2026-09-23, most recently 2026-09-08; nine Form 144 proposed sales since 2026-03-27 (2026-08-07 the latest). Transaction codes were not itemised this run — the Form 4 bodies were not opened, so no buy/sell direction is claimed.

13D/13G: no activist and no strategic block; the recent filings are passive 13Gs and the Vanguard 13G/A above.

Short interest (Equibles/FINRA): 6,318,184 shares at the 2026-08-31 settlement, +1,070,697 from 2026-08-14, 4.2 days to cover — about 2.4% of shares outstanding. There is no short case being expressed in the tape. (Equibles also returns a model estimate for the 2026-09-15 settlement of ~6.24m shares; that is a prediction, not a FINRA figure, and is not used anywhere above.)

---

8. Filings read, and the red-flag sweep (S3)

Three biggest risks, in management's own ordering, from the FY2025 10-K Item 1A summary table and the ordered body beneath it (pp.20–22):

1. "Property and casualty actual claim costs may exceed current reserves established for claims due to changes in the inflationary, regulatory and litigation environment" — first risk factor under "Insurance and financial services", p.21.
2. "Increases in the frequency or severity of property and casualty claims may adversely affect our results of operations and financial condition" — second, pp.21–22.
3. "Catastrophes and severe weather events may subject us to significant losses" — third, p.22.

Management's own summary table lists, in this order: complexity and uncertainty of loss cost estimates and reserves; claim frequency and severity volatility; catastrophes and severe weather. The two risks management ranks first are the two this dossier is about.

Auditor: Deloitte & Touche LLP, PCAOB ID 34. Sole critical audit matter: the reserve for property and casualty insurance claims, and the disclosed procedure includes "Performing a retrospective review, including comparing prior year estimates of expected incurred losses to actual experience during the current year to identify potential bias".

Red-flag sweep (efts.sec.gov full text, CIK 0000899051, 2023-01-01 to 2026-09-23):

No NT 10-K in the filing index; no 10-K/A or 10-Q/A of any year; the only amendments are 3/A, 4/A, 8-K/A, ARS/A and 13G/A. Clean.

Also noted from the 8-K cover pages: the common is now registered on NYSE Texas as well as the NYSE, while the preferred and the 5.100% subordinated debentures due 2053 remain NYSE-only.

---

9. Scuttlebutt (S6) — carried forward, plus this run's outside points

Carried from _clusters/cluster-insurance-2026-09-22.md §4: the 2026-09-07 Real Eisman Playbook episode's personal-lines passage — "which I've not been more negative on in my career" (Ryan Tunis) — with Allstate named as the captive-agent channel losing about 1.5 points of share a year to direct. That quote lives in the board's claim rows, not here.

Added this run, all [search-summary] and none of it primary:

• Mizuho raised its price target to $300, Outperform, 2026-09-18 — i.e. after the August catastrophe release.
• Zacks Research cut Strong Buy to Hold, 2026-09-17.
• Consensus Hold, average target $266.74. (This closes the analyst-coverage-count gap the cluster file flagged as a cheap fix — it cost one search, inside S6.)
• Q2 2026 adjusted EPS of $8.99 beat consensus and revenue of $18.6bn beat — the quarter itself was well received.

The product test and the expert call stay ⏳ OPEN (user).

---

10. Evidence that is thin, and access limits

• One transcript, not two. The Q1 2026 call was not read; the Equibles budget is shared across three names in this slot. The Q2 call was chosen because it is the only one that post-dates the reserve-release step-up.
• Form 4 transaction codes not itemised. 172 filings were counted and dated; none were opened, so no insider buy/sell direction is asserted.
• No shelf read. No S-3 appears in the recent index, but absence is not proof and no S-3 search was run beyond the index.
• No Investor Day or supplement. The quarterly investor supplement, where the segment-level loss-development tables live, was not read; the ratio tables used here are the 10-Q's own.
• Next-earnings date is derived, not scheduled. Equibles' IR calendar has no upcoming ALL event (last captured 2026-09-14); ROIC.ai's calendar needs a paid plan. Q3 results are placed at 2026-10-28 to 2026-11-05 from the 2024-10-30 and 2025-11-05 precedents. The September monthly release (~2026-10-15) is the earlier and more informative date and its cadence is well established.
• The $9.5bn deployable-capital figure is an analyst's, spoken in a question on the Q2 call and not confirmed by management in the turn that followed. It is not used in the model.
• No Glassdoor, G2 or channel checks. S6 is carry-forward plus search summaries.