ASIC — Ategrity Specialty Insurance Co Holdings

Ategrity Specialty Insurance Company Holdings (NYSE: ASIC) — Deep-value report (REFRESH)

Analysis date: 2026-09-01 · Price: … · Market cap: ~…M · Shares outstanding: 47,926,928 (June 30, 2026) · BVPS: … Mode: REFRESH over the reference thesis from 2026-07-24 (39 days). deep_delta.py triage: 2 new filings, 4 unverifiable falsifiers. New primary sources: 10-Q Q2 2026 (filed August 13, 2026, accession 0002040491-26-000045) and 8-K from July 29, 2026 + Exhibit 99.1 (Q2 2026 earnings release) — both downloaded in this session, not previously in the local file. Legacy sources (not re-read, used via the 07/24/2026 report): 10-K FY2025, 424B4, DEF 14A 2026, 8-K Feb/Apr/Jun/Jul 2026.

Executive summary

What I inherit unchanged from the July 24 analysis. The business description (E&S insurer for SMBs, “productionized underwriting” model, 67/33 casualty/property mix, 48 states, distribution concentration 46.5% GWP on three wholesale brokers), the governance structure (ZFSG/Stuart Zimmer controls 83.7% of votes, independently confirmed by yfinance at 81.4% heldPercentInsiders), the reinsurance profile (~80% quota-share to Ategrity Limited Bermuda, 1-in-250-year PML …M), the A.M. Best “A-” Outlook Positive rating, the board composition (7 directors, including Robert C. Merton), and the ZIS loan’s contractual structure (…M, 5.5%, due April 2032, partly collateralized with ASIC shares held by ZFSG). Nothing in the new filings contradicts these elements. What I fully re-derive: normalized earning power, quarter quality, affiliate exposure and valuation.

What changed, and why it matters. The second quarter of 2026 shows, on the surface, the best set of figures in the company’s history: gross written premiums …M (… YoY), combined ratio 85.9% (from 88.9%), underwriting income …M (…), net profit attributable to shareholders …M (…), diluted EPS … versus a consensus of … [8-K July 29, 2026, Ex. 99.1]. Beneath the surface, however, the three components of the improvement behave differently, and the difference is the thesis:

1. The combined ratio improvement is entirely on expenses, not underwriting. The expense ratio fell 350 bp (31.0% → …%), but the loss ratio rose 50 bp (58.0% → 58.5%). And the loss ratio reported in Q2 2026 benefits from favorable prior-accident-year development of …M — nonexistent in Q2 2025, where development was exactly zero [10-Q Q2’26, MD&A; press release]. On a current-accident-year, comparable basis, the loss ratio went from 58.0% to 59.3%, i.e. a 130 bp deterioration, not a 50 bp rise.
2. Nearly half of the expense improvement is reclassified fee income, not cut costs. The company reports “operating expenses, net of fee income” falling from 12.4% to 9.5% of net earned premiums. But fee income, which is subtracted in the numerator, rose from …M to …M — from 1.75% to 3.02% of net premiums, i.e. +127 bp. Of the 290 bp of “operating improvement,” 127 bp (44%) is added revenue, not reduced expense. On a gross basis, operating expenses rose 15.8% YoY, versus … net earned premiums — real leverage, but roughly half of what was communicated.
3. 40% of the quarter’s gross profit comes from mark-to-market on the affiliated hedge fund. Of …M pre-tax income, …M is realized and unrealized investment gains, almost entirely from the “Absolute Return Utility & Infrastructure Fund,” managed by Zimmer Partners LP [10-Q Q2’26, realized/unrealized gains table]. At the same time, recurring income from that fund collapsed 91.3%, from …M to …M. The fund has practically stopped producing current income and has become exclusively a source of marks.

Normalized earning power, however, has grown real and substantially. Excluding investment gains, H1 2026 profit is …M, i.e. …M annualized; the run-rate from Q2 alone is …M. Adding a normalized contribution from the affiliated portfolio (~4% net of fees and tax on …M) and applying a haircut for loss-ratio drift, the normalized owner earning is ~…M, versus …M in the July analysis. This is the most important factual move in the refresh and it’s favorable to the thesis.

But the valuation ran ahead of the facts. The price rose from … to … (…) in five weeks, over a period in which book value per share rose from … to … (…). P/B rose from 1.74x to 1.88x, while the closest comparable (Skyward, SKWD) fell from 2.13x to 2.01x. The discount to the peer group, which was the explicit engine of the July thesis, closed from 18% to 6.4% — two-thirds of it has vanished. The P/E × P/B product, criterion 3 in the tracker, went from 20.7 (below the Graham threshold of 22.5) to 23.0 — above the threshold. The mechanical falsifier “the discount hasn’t closed” still “holds” only because its threshold was set at SKWD’s July P/B; economically, it’s practically exhausted.

Estimated value. Triangulation across five models gives a range from … (bear, residual-income) to … (bull), with a base of …/share (residual-income model, normalized ROE 14%, g 6%, r …%) — i.e. … versus price. Monte Carlo on distributable earnings gives a median intrinsic value of … and an undervaluation probability of only …%. Broker consensus (…–…) is more optimistic than I am by 8…, and the current price is practically equal to the low end of the consensus target.

Verdict: NEUTRAL — hold/watch, not add. Downgraded from “INTERESTING” (the tracker’s GBL score, 19.5/30, calculated at … on 07/24/2026). Not because the business has deteriorated — premium growth is real, expense leverage is partly real, and normalized earning power has risen 39% versus the July estimate. But for three reasons that add up: the margin of safety has vanished at …; the quality of the combined ratio improvement is weaker than the headline (the loss ratio is moving in the wrong direction, exactly when retention rises from 70.0% to 74.2% of gross premiums); and the affiliate exposure, which the July report set as an explicit re-evaluation trigger (“ideally after the affiliated hedge fund and the ZIS loan exit the portfolio or shrink”), has grown from 46.5% to 48.5% of equity, with …M of new capital injected into the fund in H1 2026 versus …M spent on treasury share buybacks.