SKWD — Skyward Specialty Insurance Group, Inc.

Deep-value analysis (REFRESH): Skyward Specialty Insurance Group, Inc. (SKWD)

Analysis date: 2026-09-04 · Price: … · Shares: 44,396,493 outstanding / 46,605,142 fully diluted (10-Q Q2 2026) · Real market cap: …M · Sector: Financial Services / Insurance — P&C specialty (E&S + Lloyd’s) · Exchange: NASDAQ · Reporting currency = price currency = USD.

REFRESH regime. The reference thesis is from 2026-07-28 (verdict MONITOR at … GBL 61.0%, 18/30). Mechanical triage requested a revisit for 3 new filings (10-Q Q2 2026 filed 2026-08-07, 8-K results 2026-08-04, 8-K CFO transition 2026-08-20) and for one falsifier that remained unverifiable. What I inherit without re-derivation: the business description and moat structure (updated, not rewritten), the pre-2026 capital-allocation history, the Apollo deal architecture, the valuation framework (residual-income/EPV/multiples-type models, not DCF on CFO−capex), and the two pre-mortem scenarios tied to integration and the property market. What I fully re-derive: the owner-earnings bridge, all 5 valuation models, the Monte Carlo simulation, the Quality of Earnings chapter (absent from the prior report), the CEO profile on the Thorndike grid (absent from the prior report), the accounting red flags, and the verdict.

Primary sources this time: 10-Q Q2 2026 (SKWD_skwd-20260630.htm, CIK 0001519449, accession 0001519449-26-000054), the Q2 2026 earnings release (Exhibit 99.1 to the 8-K from 2026-08-04), the 8-K from 2026-08-20 (CFO transition), 10-K FY2025 (filed 2026-03-02), 10-Q Q2 2025 (balance-sheet comparative), deterministic data pack deep_data_pack.py (SEC companyfacts + yfinance, 2026-09-04), mc_dcf.py (20,000 scenarios), GBL tracker, web research brief 2026-09-04.

Market-data calibration note — important. Both the data pack and the delta pack, as well as yfinance, use 40,543,065 shares (the pre-Apollo-acquisition count). The actual count at 30.06.2026 is 44,396,493. Verified consequences: market cap isn’t … but … (…), and the “FCF yield 17.1%” from the data pack becomes 15.6% even before discussing whether that FCF means anything at an insurer (it doesn’t — see the valuation chapter). The data pack’s P/B (2.03x) is correct, since it’s computed as price/BVPS, not market cap/equity. Every per-share figure in this report uses the filed share count, not yfinance’s.

Executive summary

The second quarter of 2026 was the thesis’s first real test, and the result changes the July report’s conclusion. On July 28 I wrote “quality company, not cheap, watchlist, accumulate below …-52.” Three things have happened since, and none of them is the price move (the stock went … → … → … i.e. noise).

First: the combined ratio didn’t deteriorate, even though management itself had guided deterioration. FY2026 guidance, issued in December 2025 and reaffirmed in April, says combined ratio 90.5…. The first half delivered 89.5% consolidated (87.6% ex-catastrophe), identical in Q1 and Q2, with the US segment at 87.9% and Apollo at 91.3%. The company beat its own guidance by 1.0–2.0 points for two quarters running and did not raise guidance. The falsifier “the property market softens and margin goes below threshold” didn’t trigger.

Second: underwriting discipline is visible in the numbers, not just in press releases. Of the nine underwriting divisions, five CONTRACTED in H1 2026 — Energy Solutions …, Global Property …, Captives …, Professional Lines …, Transactional E&S … — exactly the lines where E&S pricing is softening. Growth came from Global Agriculture (…), A&H (…), Specialty Programs (…) and Credit & Surety (…). In July I wrote that the “execution moat” was unproven through-cycle; this is the first direct evidence that the team is withdrawing capital from lines where price no longer pays for risk. Not complete proof (one half-year), but more than I had.

Third, and most relevant for capital allocation: buybacks started, contrarian. From October 2024 (when the …M program was authorized) to May 2026 — twenty months during which the stock climbed toward … — zero shares were bought back. In June 2026, with the stock averaging … (~8% above the 52-week low of …), the company bought 222,635 shares for …M — all in a single month, with April and May empty. On July 15, 2026 the board doubled the authorization to …M. In the SAME quarter the company repaid …M of Tranche A of the term loan (June 26), six months after a …M acquisition. Debt repayment plus buybacks at the low, simultaneously: this is the behavior Thorndike looks for, and it’s new versus July.

What got worse, and must be said just as clearly: cash conversion. Operating cash flow for H1 2026 was …M versus …M in H1 2025 — flat to zero — even though net profit grew 22.1% and revenue 48.9%. Premium receivables jumped from …M (June 2025) to …M (June 2026), and DSO rose from 84.5 to 126.8 days. A substantial part is structural (Lloyd’s settlement cycle is notoriously slow, and Apollo brought …M of receivables on January 1), but not all of it. This is the main signal to watch at the October 29 report.

The owner-earnings bridge (declared, not implicit): for an insurer, CFO − capex isn’t owner earnings, it’s float growth. The …M of “FCF” in the data pack is overwhelmingly premium collected that hasn’t yet turned into claims paid — policyholders’ money, not the shareholder’s. My base is normalized owner profit: …M (…/diluted share), built from underwriting income plus net investment income, minus interest, minus recurring intangible amortization, with the combined ratio normalized to 92% (versus 89.5% realized) and VOBA amortization (non-recurring) excluded. Full detail in the valuation chapter.

Valuation: five models give a range from … (DCF with combined ratio at 95% and 12% cost of capital) to … (DCF with CR 90% and r 9.5%), with the median at … and the base-case DCF at … (…/share). Monte Carlo over 20,000 scenarios gives a median of … and an undervaluation probability of …%, with a P10–P90 range from … to …. Consensus of 11 analysts (yfinance, 2026-09-04) is … range …–…: my base value of … lands within 1.4% of consensus, a good external check on the bridge, not a coincidence — both start from the same guidance figures.

Verdict: GRADUAL ACCUMULATION, upgraded from MONITOR. Not because the stock is cheap in the Graham sense (P/B 2.03x, tangible P/B 2.99x aren’t cheap by any reading), but because three of July’s conservative assumptions proved too conservative (margin, discipline, capital allocation), and the price is 5.5% below then. Position size, however, must be limited by the range’s width: P10 at … isn’t a theoretical tail, it’s the scenario where cash conversion doesn’t repair and the combined ratio climbs toward 93… in a soft market.