Universal Insurance Holdings, Inc. UVE

UVE — Universal Insurance Holdings, Inc. — Deep-value analysis (REFRESH)

Date: 2026-09-04 · Price: … (previous close …) · Market cap: ~… · Sector: Financial Services / Insurance – Property & Casualty · Exchange: NYSE (US company; no symbol collision with BSE — confirmed in tracker, exchange=NYSE)

This analysis is strictly informational/educational, not investment advice. Figures come from SEC EDGAR (10-K FY2021–FY2025, 10-Q Q1 2026 and Q2 2026 — new, 8-K 05.06.2026 and 18.06.2026 — new, DEF 14A 24.04.2026 — new), from the deterministic data pack (XBRL companyfacts + yfinance), and from the research brief; the source is noted at each key figure.

What I inherit and what I re-derive (REFRESH regime note)

The mechanical deep_delta.py triage flagged a single "BROKEN" falsifier: shares > 29000, with a measured value of 27,830,355. The falsifier is spuriously broken, from a units mismatch, not a change in facts. The threshold was written in the 08.08.2026 thesis in THOUSANDS of shares (29,000 thousand = 29.0 million), but teza_deep.py:380 maps the shares metric to yfinance's sharesOutstanding, which is a RAW share count. The actual comparison was 27,830,355 > 29,000 — true by construction, regardless of what the company does. Reality is the exact opposite of the falsified claim: shares fell from 28.3M (2024-09) to 27.83M (2026-06), and buybacks continued uninterrupted in H1 2026 (331,550 shares, …M — 10-Q Q2 2026, Note 8). The capital-allocation thesis HOLDS. I flag the bug explicitly, since otherwise the next triage run will trigger it again, endlessly, on any ticker.

What the triage did not flag, and which is actually the real reason this report is being rewritten: the delta pack says "new filings: none," but the local SEC file set had stopped at the Q3 2025 10-Q, while the company had in the meantime filed 10-Q Q1 2026 (29.04.2026) and 10-Q Q2 2026 (30.07.2026), plus two material 8-Ks in June. edgar_10k_downloader.py only downloads 10-Qs within the fiscal year of the last 10-K (FY2025), so 2026 quarters weren't pulled in. I downloaded them specifically and indexed them with filing_text.py. Almost everything new in this report comes from these three documents, not from the price move (… since the reference analysis — irrelevant and not used as an argument).

I inherit unchanged, from the 08.08.2026 report: the vertically integrated structure and moat description (ERA, Alder, BARC, Clovered, ~9,500 agents, 92.5% retention); the 2020–2025 buyback history with average prices; the 2020–2025 reserve adverse-development history; the fixed-dividend analysis; the absence of related-party transactions; the Mangrove/VIE captive structure; the 2020 investment-gains one-off history.

I fully re-derive: the owner-earnings bridge and normalized-earnings base (where the old report contained a construction error), all five valuation models, Monte Carlo, the last-4-quarters balance-sheet chapter, the Quality of Earnings verdict, the CEO profile (the April 2026 proxy overturns the previously used insider-ownership figures), and the red-flags list.

Executive summary

Universal Insurance Holdings is a vertically integrated insurance holding company, writing homeowners policies through UPCIC and APPCIC in 19 states (also licensed in Tennessee, with a rate filing pending), of which Florida represents 72.9% of direct premiums written in H1 2026, but only 47.1% of total insured value (10-Q Q2 2026, Overview). August's reference thesis said: a real regulatory catalyst, conservative capital, but peak-cycle EPS and no margin of safety (median MOS …). Three new facts from the 2026 filings materially change the quantitative part of that thesis, and a fourth is a construction error in the old report that I correct here.

New fact #1 — reinsurance got 12.1% cheaper, for more coverage. The 2026-27 program, effective June 1, 2026, has a ceded written premium of …lion, versus …lion under the prior program (−…M), while the single-event coverage tower INCREASED by ~…M, to … billion, and the statutory first-event retention stayed unchanged at …M (10-Q Q2 2026, "Reinsurance"). An additional …M of forward capacity extends into the 2027-28 treaty period, of which …M is under the FHCF layer. This directly overturns Scenario 3 in the old report's pre-mortem ("reinsurance costs already rose 10.9% in a benign year"). And, crucially, the benefit is almost entirely IN THE FUTURE: reinsurance costs are recognized pro rata over the June 1 – May 31 period, so in H1 2026 ceded earned premium was still 31.8% of direct earned premium, HIGHER than 30.9% in H1 2025. The contracted ceded written premium (…M) represents 28.5% of annualized direct earned premium — a gain of ~3.3 points of combined ratio not yet in any reported result.

New fact #2 — reserve adverse development was ZERO in H1 2026. "During the three and six months ended June 30, 2026, there was no prior year development" (10-Q Q2 2026, Note 6). This was exactly the metric August's report had flagged as the signal to watch. But disciplined restraint is mandatory: at UVE, development is booked seasonally, in H2. In 2025, H1 had … thousand of development, Q3 added …M, and the full fiscal year closed at …M — meaning ~…M was booked in Q4, at the year-end actuarial review. Zero in H1 2026 is therefore consistent with the pattern, not proof. The real test comes on 22.10.2026 (Q3) and at the February 2027 10-K. What IS operational proof, from the April 2026 proxy: under Downes's leadership of the claims and litigation departments, in 2025 "Universal experienced a 31% reduction in incoming lawsuits and a 60% reduction in active lawsuit inventory" — the pre-reform tail is dying out, and payments on prior years fell from …M (H1 2025) to …M (H1 2026).

New fact #3 — refinancing risk is closed, at a cost of 212.5 basis points. On June 16, 2026 the company issued …M of 7.75%-coupon senior unsecured notes due June 30, 2031 (private placement), and on June 17 fully redeemed the 5.625% notes due November 2026, at an extinguishment loss of … thousand (8-K 18.06.2026; 10-Q Q2 2026, Note 7). KBRA assigned UVE an issuer rating of BBB, stable outlook — the holding company's first public credit rating. The incremental cost is …M/year pretax (~…/share after tax), plus …M of issuance costs amortized over five years. The new covenants (debt/capitalization ≤ 40%, cash ≥ 12 months of interest, dividend lock-up on a default event) are comfortable: the current ratio is 13.6%.

Correction #4 — the normalized-earnings base in the old report was built incorrectly. Model 2 there took a normalized combined ratio of …% and directly declared a normalized EPS of … The problem: at UVE, the combined ratio applies EXCLUSIVELY to net earned premium, and below that line sits …lion per year of revenue with no combined ratio at all — net investment income (…M), reinsurance brokerage commission earned by BARC (…-60M), MGA policy fees (…M), and other revenue (~…M). Exactly the vertically integrated part of the business, the qualitative core of the thesis, had been excluded from the quantitative side. Rebuilding the bridge in dollars and checking it against fiscal 2025 (where it reconciles to ±…M), mid-cycle normalized owner earnings come out at …lion, i.e. normalized EPS …, not … — with a catastrophe load of 7.0% of net premium already deducted.

Estimated value and verdict. Triangulation across five models gives an extremely wide range: … (bear) to … (bull), with a median of …. Monte Carlo over 20,000 scenarios gives a median intrinsic value of … and a median MOS of ** …**, with an … undervaluation probability. The divergence between the triangulated median (…) and the Monte Carlo median (…) isn't an inconsistency: Monte Carlo simulates exclusively the DCF branch, while the Earnings Power Value anchors (…) and P/B on normalized ROE (…) — which assume no growth — pull the median back toward the market price. The two no-growth anchors land practically on price and on the sole active analyst's target (… Piper Sandler, 27.07.2026): the market is paying today exactly UVE's normalized earning power, with no growth. The verdict remains INTERESTING, but with meaningfully more conviction than in August (median MOS moved 31 points, from … to …) and with an explicit, dated trigger: the Q3 report on October 22, 2026. If that quarter again shows neutral-or-favorable development AND no material catastrophe, the last substantive falsifier falls, and the base case migrates toward the bull model.