ASIC — Ategrity Specialty Insurance Co Holdings

Deep-Value Analysis — ASIC (Ategrity Specialty Insurance Company Holdings, NYSE)

Date: September 19, 2026 · Price: USD … · Market cap: USD 1.323 bn · Sector: Financial Services / Insurance — Property & Casualty · Single currency: USD reporting / USD price

Run type: mechanically triggered re-analysis. The thesis written on September 1, 2026 carried nine load-bearing claims with falsifiers; the check on September 17 reported one as broken — dso. This run’s job is to establish whether the hypothesis behind it actually broke, and only then to rebuild the analysis.

Executive summary

The answer to the question that triggered this run, first of all: the dso falsifier does not signal any deterioration. It is a calibration artifact — the threshold was written on one scale, and the metric is measured on another. The September 1 thesis stated “DSO oscillates in a 2.2-day band (48.4 → 50.6 → 49.1)” and attached the threshold dso > 65 to it. That 49.1 is premium receivables (USD 111,638 thousand) over the quarter’s gross written premiums (USD 206,762 thousand) — the correct definition for an insurer. The mechanical metric in the thesis pipeline, however, computes receivables / total revenue × 91, and “receivables” in yfinance’s balance sheet is the sum of premium receivables and reinsurance recoverables: 111,638 + 188,156 = 299,794 thousand USD, divided by the quarter’s total revenue of 148,484 thousand, times 91 = 183.7 days — exactly the value the falsifier reported. The reinsurance recoverable is not a trade receivable: it is the asset corresponding to ceded losses, i.e. the portion of the company’s own reserves that someone else will pay. It has no connection to “revenue without collection,” which is the object of the O’Glove test the threshold was meant to protect.

The decisive check is the history. Reconstructed from the quarterly filings, the mechanical metric reads: 218.7 (06/30/2025) → 190.6 (09/30/2025) → 167.9 (12/31/2025) → 168.0 (03/31/2026) → 183.7 (06/30/2026). It has never been below 65 since the company went public — so the threshold was not “breached,” it was impossible to satisfy from the day it was written. Moreover, over the 12-month window the trend is improving (218.7 → 183.7, i.e. …), not deteriorating. The falsifier only triggered now because it required two consecutive quarters, and the 03/31/2026 balance sheet only fully appeared in the mechanical source recently. On the correct scale — premium receivables / gross written premiums — the series is 48.4 → 51.1 → 49.1 days, a 2.7-day band, and the year-over-year growth in receivables (…) tracks the year-over-year growth in gross written premiums (…) within 1.8 percentage points. Verdict on the broken hypothesis: calibration noise, not structural deterioration. The old thesis’s claim is rewritten identically, with the threshold moved to 260 days — a level that, 19% above the historical observed maximum, would only trigger on a real explosion of either uncollected receivables or ceded losses.

But the re-analysis found something else, in another part of the same balance sheet, and it is this report’s single most important number. The loss ratio calculated on a gross basis — direct losses incurred over direct premiums earned — deteriorated by 863 basis points in the second quarter: 103,468/151,462 = 68.3%, versus 74,382/124,624 = 59.7% in Q2 2025. Over the half-year, the deterioration is 683 bp (62.3% versus 55.4%). The net ratio, the only one the investor sees in the combined ratio, barely moved (58.5% versus 58.0%). The difference was absorbed entirely by reinsurers: ceded losses rose 54.2% (36,965 versus 23,970) on ceded earned premiums that were practically unchanged (37,687 versus 37,696), giving a ceded-program loss ratio of 98.1% in Q2 2026, versus …% in Q2 2025. In a single quarter, reinsurers underwrote ASIC at a loss. Calculated over four quarters, however, the same ratio is roughly 50% — so Q2 is an isolated data point, most likely a property loss event that hit the ceded layer, not a trend. That turns it from a certain penalty into a pre-mortem scenario with its own falsifier, and is why I do not cut the valuation base by the full effect.

What actually changed versus the September 1 analysis. No new filing — Q3 2026 publishes October 21. The price changed: … → …, … in eighteen days, on a P/B that rose from 1.895× to 1.992×. The original July thesis’s engine was the discount to the comparable group; by September 1 it had narrowed to 6.4% versus Skyward Specialty (P/B 2.01×). Today, with SKWD still at roughly 2.0× book value, the discount is zero. The engine is fully spent. And one of my own assumptions changed: I normalize owner earning power on the half-year, not on annualized Q2, which lowers distributable owner earnings from 52 to 46 million USD.

Estimated value. Five models triangulated on the same assumptions (OE USD 46 mn, g₁ = g_t = 6%, r = …%, zero net debt, 47.927 mn shares): EPV Greenwald … · DCF bear … · DCF base … · Graham Number … · DCF bull …. Monte Carlo on 20,000 scenarios: median intrinsic value …, median margin of safety …, probability of undervaluation …%, probability of a margin above 30% only 7.1%. Of the … points of negative discount, roughly 16 come from price and the unmodified model, and 10 from my revision of the base — stated explicitly because it is opinion, not fact.

Verdict: AVOID at …. Downgraded from NEUTRAL (September 1) and from INTERESTING (July 24). Not because the business broke — gross written premiums are growing 23.4% in an E&S market the company’s own management describes as contracting, the combined ratio is 85.9%, the balance sheet carries no financial debt, and receivables are clean. But because three things add up: the valuation discount has vanished entirely; distributable earning power is roughly 43% of reported profit, the rest being either mark-to-market on an affiliated fund or capital that must be retained to fund premium growth; and half of book equity (USD 333.8 mn, 50.2%) is tied up in exposures to the controlling shareholder, with a contractually asymmetric liquidity mismatch favoring the affiliate and fees leaking to it at an annualized rate of roughly USD 19.5 mn. At 20-… the thesis becomes interesting again; at … it does not.