ACIC — American Coastal Insurance Corporation

ACIC — American Coastal Insurance Corporation · deep-value re-analysis (falsifier triggered: dso)

Date: September 16, 2026 · Reference price: $9.14 · Market cap: $432.0 mil. (47,269,517 shares) · NASDAQ · Sector: property insurance, commercial-residential, Florida

Why this report exists. On 09.15.2026, the mechanical triage ran the 6 falsifiers of the thesis written on 08.16.2026 and one triggered: dso, with values of 261.8 and 241.6 days against a threshold of 90. This session’s job is to answer a single question — does the hypothesis that supported the verdict still hold? — and only then redo the full analysis.

Primary sources: 10-Q Q2 2026 (filed 08.06.2026), 10-Q Q1 2026 (05.08.2026), 10-K FY2025 (09.03.2026), 10-K FY2024, SEC XBRL companyfacts (CIK 0001401521, direct extracts for the receivables, premium and reserve series), the automated data pack, the research brief (web brief — treated as a deposition, two figures corrected below), Q2 2026 earnings call transcript (08.05.2026), GlobeNewswire press release 09.14.2026, the thesis file, the Monte Carlo simulation, GBL tracker (score dated 09.06.2026).

Two corrections to the research brief, before anything else:

1. The brief writes “guidance for fiscal year 2026: profit $85-100 mil.,” implying net profit. The company’s original wording is “earnings before income tax of $85 million to $100 million” — profit BEFORE tax. The difference is $22 million a year and changes the valuation base by 32%. The correct figure, after the H1 2026 effective tax rate of 24.0% ($13,006 thousand / $54,156 thousand), is $64.6-76.0 mil. net profit, midpoint $70.3 mil.
2. The brief attributes to a “compiled SEC 8-K source” an FHCF contract effective 06.01.2026 with 90% coverage, mandatory layer ~$571.5 mil. There’s no mention of FHCF anywhere in the Q2 2026 10-Q, the FY2025 10-K, or the reinsurance program description (Note 6 and the Reinsurance section). The filing’s description of the program is explicitly private occurrence-based, with exhaustion at $1.68 bn on the first event. I don’t use the FHCF figure anywhere in this report; if it exists, it isn’t in the public file I read.

---

Executive summary (1 page: thesis, estimated value, verdict)

The answer to the question that commissioned this report: the falsifier triggered on a metric the thesis wasn’t measuring. The mechanical checker computes dso as yfinance receivables / quarterly revenue × 91. For ACIC, yfinance’s “Accounts Receivable” line is $237.3 mil. as of 06.30.2026 — that’s premiums receivable $119,915 thousand PLUS reinsurance recoverable $117,435 thousand, a sum that reconciles to the last dollar with the 10-Q balance sheet. The reinsurance recoverable is a claims-recovery asset, not a revenue collection; including it inflates the ratio 2-3x. Consequence: the metric was already above 90 in every quarter yfinance retains — 176.0 (09.30.2025), 209.5 (12.31.2025), 241.6 (03.31.2026), 261.8 (06.30.2026). The falsifier was triggering from the day it was written. Between 08.16.2026 and today no new filing has appeared (Q2 was filed 08.06.2026; Q3 comes 11.04.2026), so zero new information caused the trigger.

On the metric the thesis actually measured, the verdict is more nuanced — and more interesting. I reconstructed the prior author’s definition from his own figures: premiums receivable / TTM gross earned premiums × 365. It reproduces the cited series exactly (27.7 for FY2023, 26.6 for FY2024, 39.7 for FY2025, 72.9 for TTM as of 06.30.2026 — all four match to a decimal). At 72.9 days, the 90 threshold isn’t reached: the falsifier is at 81% of the threshold. But the claim it was protecting wasn’t “DSO stays below 90,” it was “the receivables growth remains isolated — hasn’t worsened into a structural pattern.” That claim is falsified by the evidence, even if not by the number: 26.6 → 39.7 → 44.6 → 72.9 are four consecutive rising observations, and receivables relative to unearned premiums went from 13.9-19.2% (2024 – mid-2025) to 28.2% / 29.8% / 35.7% (Q4 2025, Q1 2026, Q2 2026). It’s not an isolated quarter. It’s a level shift over three quarters, plus a $43.2 mil. jump concentrated in Q2 2026.

Structural deterioration or noise? My verdict: a real symptom of a soft market, with near-zero credit risk, amplified by the worst possible measurement date. The reasons, each with its own figure: (a) the credit-loss allowance on premiums receivable is zero as of 06.30.2026 ($0; $26 thousand a year earlier), and the Note 17 roll-forward shows no write-offs; (b) the policy stated in the 10-K is a policy-level test whereby the receivable can’t exceed the associated unearned premium — at $119.9 mil. receivables against $335.5 mil. unearned premiums, coverage is 2.8x, and a non-payer is canceled with the unearned premium released, so loss-given-default is near zero; (c) the half-year’s operating cash flow is POSITIVE $52.8 mil., so the company generated cash after absorbing the $49.5 mil. receivables build; (d) June 30 is the balance-sheet date right after the June 1 renewal, when the entire portfolio is billed and uncollected — historically, the balance collapses in Q3 ($18.6 mil. at 09.30.2024, $30.3 mil. at 09.30.2025). Against: there’s no explanation in any filing or on the analyst call, and the context — net price down 24%, new and renewal policies +11.3%, policies in force +1.2% — describes exactly a firm defending volume with wider commercial terms. The deciding test is the 09.30.2026 balance, published November 4.

What actually changed versus the prior analysis. One material thing, and it’s positive: on 09.14.2026 the company announced a restructuring of the exclusive agency agreement with AmRisc, effective 07.01.2026. The old contract had a fixed expiration of 01.01.2029 — i.e. the firm’s largest single-point-of-failure risk (97% of the commercial-lines business, 100% of condominium revenue, plus premium collection) had a contractual cliff 27 months away. The new agreement has no expiration date, auto-renews annually and requires 48 months’ notice plus a minimum four years from notification. The risk hasn’t disappeared, but its horizon has moved from 2029 to “at least 2030, with four years’ notice.” Otherwise: Q2 2026 is the same quarter the prior report read, the price fell from $9.86 to $9.14 (−7.3%), and KBRA raised AmCoastal’s financial-strength rating to A from A− and the issuer rating to BBB from BBB− (07.21.2026, Note 18, subsequent events).

Estimated value. The stated owner-earnings base is $70.0 mil. — not CFO − capex, which over the trailing 12 months gives minus $31.0 mil. from post-Milton claims settlement. The external check the method requires: the 3.5-year average of operating cash flow minus capex (2023: −$136.0 mil.; 2024: +$243.5 mil.; 2025: +$71.0 mil.; H1 2026: +$52.8 mil.) is +$66.1 mil./year — within 6% of the stated base, so $70.0 mil. is confirmed by cycle-level cash, not invented. The implied owner-earnings yield, 16.2%, sits between the trailing yield (21.6%, P/E 4.64) and the published forward yield (11.9%, forward P/E 8.39) — consistent with what the market sees, unlike the “−7.2% FCF yield” in the data pack, which is pure calendar artifact. The five-model triangulation gives a margin-of-safety range of −21.6% to +129.2%, median +31.6%; Monte Carlo across 20,000 scenarios gives median intrinsic value $12.96/share, median MOS +41.8%, undervaluation probability 87.0%.

Verdict: INTERESTING — small accumulation maintained (1-2% of portfolio), explicitly conditional on the November 4 report. The thesis doesn’t die. The falsifier that triggered was measuring something else; and on the business fundamentals, the only material change in the last month is a structural improvement (AmRisc). But the claim “receivables remain an isolated signal” was wrong and I withdraw it: the level shift is real, has lasted three quarters, and has no published explanation. Sizing stays as before — small, assuming a hurricane year, not a quiet one — and I add a verifiable condition: if premiums receivable as of 09.30.2026 doesn’t fall below ~$50 mil., the seasonal explanation dies and the thesis must be rewritten without it.

---