HCI Group, Inc. HCI

Deep-value analysis — HCI Group, Inc. (NYSE: HCI) — REVISIT (refresh)

Analysis date: 2026-09-05 · Reference price: … · Market cap: …M · 12,438,867 shares (yfinance, 2026-09-05) REFRESH regime. Reference thesis: rapoarte/deep/2026-08-14-deep-HCI.md (22 days). Delta pack: rapoarte/deep/delta-HCI-20260904.md. Reason for revisiting: a single unverifiable falsifier — year-on-year revenue growth, for which the mechanical triage needed two consecutive quarters and had found only one.

---

Executive summary

What I inherit and what I re-derived. The regime is refresh, so I state up front exactly what I did not rebuild. Zero new SEC filings since the thesis date: the last one is the Q2 2026 10-Q from August 7, 2026, the same one the August 14 report relied on; the next reporting is scheduled for November 5, 2026 (tracker). I therefore inherit, without re-verification: the business description and five-segment structure, the 2022–2026 capital-allocation history, the O'Glove receivables/reserves/debt tables over 3-5 years, Paresh Patel's Thorndike profile, and the FCF bridge with normalized owner earnings of …M. What I re-derived, in a targeted way, is only: (1) the broken falsifier — revenue growth over two consecutive quarters, from the Q2 10-Q's income statement; (2) the breakdown of that growth, because the "holding" figure hides where it comes from; (3) the sum-of-the-parts, since XZO moved; (4) the seasonal and rate context, where the only genuinely new information is; (5) three factual corrections — two to the research brief, one to my own August report.

The falsifier is now verified and holds — but not for the expected reason. The required metric was year-on-year revenue growth over two consecutive quarters. From the consolidated income statement (10-Q Q2 2026, "Total revenue" line): Q1 2026 …M vs Q1 2025 …M = …, Q2 2026 …M vs Q2 2025 …M = …. Both well above my base-growth assumption g1 = 6%. The falsifier not only holds, it shows the assumption is conservative. The problem is that the roughly 11% isn't all earned. Gross earned premiums — the real engine of an insurer — grow much more slowly and decelerate within the half: Q1 2026 …, Q2 2026 …. The gap between … total revenue and … gross premiums comes from three places, one of which is problematic: investment income (… in H1, legitimate), Exzeo revenue from third parties (from …M to …M in H1, legitimate and desired), and the drop in the cession rate to reinsurers, from …% to 31.8% (MD&A Q2 2026, verbatim). The last one isn't growth — it's risk retained on the company's own balance sheet.

The quantification, because it's this refresh's new conclusion. If the H1 2026 cession rate had stayed at …% (the H1 2025 level), net earned premiums would have been …M instead of …M, and the half's total revenue …M instead of …M. Revenue growth would have been …, not …. That is, 2.5 percentage points of 11.7 — a fifth of the growth — are bought with hurricane exposure, not underwriting. This is exactly the mechanism I anticipated in August's red-flags chapter as a "2027 risk flag" tied to the Cayman captive Fortex (licensed in March 2026, with the stated purpose "to selectively retain risk and reduce the cost of third-party reinsurance"). It's no longer prospective. The cession series: 35.2% (2023) → 37.5% (2024) → …% (2025) → 31.9% (Q1 2026) → 31.7% (Q2 2026). Four consecutive quarters of decline.

What changed outside the company — and it matters more than what changed inside it. Two things, in opposite directions:

1. The 2026 hurricane season has, so far, been empty. At September 5, 2026 — five days before the statistical peak of the season — the Atlantic produced two tropical storms (Arthur, Bertha) and zero hurricanes, both making landfall in Louisiana, none in Florida. NOAA maintains a below-normal seasonal forecast; Colorado State (August) models 9 named storms, 4 hurricanes, 1 major, with major-hurricane US landfall "well below the long-term average," against a backdrop of an El Niño suppressing activity. Consequence for HCI: a high probability of a second consecutive year without catastrophes, so a very good reported 2026. Consequence for valuation: none. The normalized …M pre-tax cat load remains exactly what it was — a normalization, not a weather forecast. Two gift years don't turn a gift into a rent; they just make reported EPS even more unrepresentative. What changes is the distribution over the pre-mortem's 24-month window: half of it is now known and quiet, which lowers the "hurricane" scenario's probability from ~35% to ~25% for the remaining window.
2. Florida rates really are falling, at competitors, right now. In August my report said FLOIR had "announced" rate relief and that "it doesn't yet appear in Q2 2026" in any negative rate filing. In the meantime the figures are public: rates falling in 51 of 67 counties in 2026, 44 insurers have filed rate decreases and another 48 filings with zero-percent changes since 2024, with big, concrete examples — State Farm … statewide, Florida Peninsula …, Patriot Select …, Heritage … in Seminole. HCI doesn't appear on the list of those who cut. That isn't good news: in a market where 44 competitors are cutting 8… while you hold your average premium flat, you either lose volume or follow. Pre-mortem Scenario 2 — "the regulator and competition take the margin back" — has moved from prospective to in progress. I raise its probability from ~40% to ~50% and confirm it as the most likely way the thesis loses money.

Estimated value, recalculated at today's price. The valuation assumptions are unchanged — I have no reason to change them, since nothing in the filings moved. What moved is the XZO price (…, from … to …), which enters the sum-of-the-parts:

Model	Value/share	MOS vs …	Δ vs 08/14
DCF bear (heavy cat load + FL rate cuts)	…	…	−0.5pp (price only)
EPV Greenwald (no growth, r=11%)	…	…	−0.6pp (price only)
Sum of parts (XZO at market + stub at 8×)	…	…	+2.0pp (XZO …)
DCF base	…	…	−1.0pp (price only)
DCF bull	…	…	−1.2pp (price only)

Median of the five: … (was …). Monte Carlo on the base model, 20,000 scenarios, with my assumptions: median …, P(undervalued) …%, but P10 = …. The central P25–P75 interval is … … … — 57 percentage points wide.

The inverse reading, which is the most useful figure in the whole report. HCI's market cap is …M. The 83.1% stake in XZO is worth, at today's market price, …M — that is, 53.1% of HCI's market cap. So the market is valuing everything else — insurance, the Greenleaf real estate, surplus notes, minus debt — at …M, for normalized after-tax insurance profit of …M: 6.1× earnings. Three weeks ago it was 6.4×. The "stub" discount widened with HCI's price absolutely flat, purely because XZO re-rated upward. It's an arithmetic improvement, not an economic one: it depends entirely on a multiple the market places on a profit that's 89% intercompany.

Verdict: MONITOR, unchanged. A small position (1…) is justifiable, accumulation below … The refresh produced no reason to upgrade or downgrade. It did produce a better signal to watch than August's: the cession rate. If at the June 1, 2027 renewal it falls below 30%, HCI has decided to fund its revenue growth with its own balance sheet, and the first encounter with a major hurricane will cost far more than 2024 did.

---