NMI Holdings, Inc. NMIH

NMI Holdings, Inc. (NMIH) — Deep-value report (REFRESH)

Generated: 2026-09-04 · Reference price: … (close 2026-09-03, the price also used by the Monte Carlo simulation on disk) · Spot 2026-09-04: … (yfinance, market cap 3.414bn / 75.218M shares) — at spot, every MOS below falls by 1.8pp · Reference market cap: ~… · Sector: Financial Services / Insurance – Specialty (private mortgage insurance)

Regime: REFRESH over the reference report 2026-07-27-deep-NMIH.md (verdict then: INTERESTING, median MOS …). Mechanical trigger: 2 new filings — 10-Q Q2 2026 (filed 2026-07-31, accession 0001547903-26-000038) and the 8-K with the earnings release (2026-07-30, 0001547903-26-000037). All 6 falsifiers in the delta pack delta-NMIH-20260903.md hold (none broke); the price move from the reference analysis is …, i.e. noise, and is not the reason for revisiting.

Primary sources: SEC EDGAR CIK 1547903 — 10-Q Q2 2026 (SEC-Filings/NMIH/NMIH_10Q_2026-07-31.txt), 10-K FY2025 (NMIH_10K_2026-02-12.txt), 10-K FY2023 (NMIH_10K_2024-02-15.txt), EDGAR companyfacts XBRL via rapoarte/deep/data-pack-NMIH-20260904.md. Market context: rapoarte/deep/research-NMIH-20260904.md (research brief, checked point by point below). Simulation: rapoarte/deep/mc-NMIH-20260903.json.

---

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

Inherited unchanged (verified the new filing didn't move it): industry structure — an oligopoly of six GSE-approved insurers regulated as such; the 2022–2025 capital-allocation history; management's identity and tenure (Pollitzer CEO since 1.01.2022, Swithenbank CFO since 1.05.2024); the reinsurance architecture as a principle; the absence of a dividend; the Wisconsin statutory accounting quirk (the contingency reserve runs through statutory income).

Fully re-derived, because the new filing or my own error invalidated it:

1. The FCF bridge and, through it, the entire valuation. The reference report used CFO − capex = …/share as the owner base, normalized down to … For a mortgage insurer this is conceptually wrong: CFO includes collected premium that must stay at NMIC against assets required by PMIERs and future claims — it isn't distributable. The correction cuts the base by ~40% and is the biggest change in this refresh.
2. The effective tax rate. Red flag #1 from the old report ("decline from 22.2% to 20.5%") is withdrawn as my own error — the actual figure from the FY2025 10-K income statement (lines 1488-1489) gives 22.18% in 2025, 22.29% in 2024, 21.95% in 2023. The rate is flat, not declining.
3. The "cheap on any multiple" positioning. At … NMIH is at 8.77x trailing profit, versus its own 5-year average of 7.37x — at the top of its own historical range, not at the bottom.
4. The reinsurance program. The Q2 2026 10-Q discloses for the first time future cessions: the 2026 QSR cedes 25% (versus 20% in 2023/2024/2025), and the 2027 QSR will cede 35%. This wasn't in the old report.
5. All quarterly portfolio, claims and capital metrics as of 30.06.2026.

---

Executive summary

NMI Holdings insures, through subsidiary NMIC, the lender's loss portion on conventional mortgages with LTV above 80%, in an oligopoly of six insurers approved by Fannie Mae and Freddie Mac. At June 30, 2026 it had …mn primary insurance in force (IIF) and …mn risk in force (RIF), across 694,273 policies and 2,223 Master Policy customers (10-Q Q2 2026, Item 2). The quarter was operationally good: record total revenue … thousand (… YoY), net profit … thousand (…), diluted EPS … (…), combined ratio 27.7% versus 28.8%, and NIW …mn — … YoY, the strongest volume acceleration in the series (10-Q, Consolidated Results of Operations and Portfolio Data). The balance sheet remains a fortress: PMIERs available assets …mn against a …mn requirement, i.e. a …mn excess and a 173.7% sufficiency ratio (up from 168.4% a year earlier); RTC …:1 versus a 25:1 cap; debt …mn against equity of …mn, i.e. 13.4% of total capitalization versus a 35% covenant.

And yet the thesis has worsened — not because of the business, but because of my own error, which this refresh corrects. The reference report treated CFO − capex (…mn ttm, …/share) as owner cash flow. For a mortgage insurer, this is an accounting illusion: of the …mn of CFO generated over the trailing 12 months, …mn was tied up purely to cover the growth in assets required by PMIERs (from …mn to …mn), and the rest sits at NMIC, which can only upstream …mn per year to the holding — the ordinary dividend paid on June 3, 2026, "representing the entire ordinary dividend capacity for the twelve-month period ending December 31, 2026" (10-Q, Liquidity and Capital Resources). The correct flow is FCFE after the regulatory capital requirement: … − … − … = …mn, which I round conservatively to …mn (…/share). The same figure emerges through the fully independent path of normalized earning power: ttm profit normalized to a 22% loss ratio = …mn, multiplied by (1 − g/ROE) at g = 4% and normalized ROE 13.5% → …mn. Two independent bridges meeting at …mn — that's the base, not …mn.

With the corrected base, triangulation across five models gives a range of … to …, median …, versus the … median from July. Monte Carlo (20,000 scenarios, mc-NMIH-20260903.json) confirms independently: median intrinsic value … median MOS …, undervaluation probability …, but only …% probability of finding a 30% margin. The 6-analyst mean target (…) implies …. Three sources converging in the … … … band — a real discount, but a thin one.

What actually worsened during the quarter, beyond my own correction: paid losses climbed to …mn over the trailing 12 months versus …mn in FY2025 and …mn in FY2024; severity jumped from 75% (H1 2025) to 88% (H1 2026), and the average amount per claim from … to …; the cure rate on new defaults fell from 93.9% (H1 2025) to 86.2% (H1 2026); persistency kept declining, 84.1% → 81.4%; and the profit commission from the quota-share treaties, a direct contributor to net premium income, fell 16% YoY (… → …mn for the half-year) precisely because ceded losses are rising. On top of all that, the 2027 QSR will cede 35% of the risk on policies written in 2027, versus 20… in recent years — a prudent risk decision, but one that cuts directly into the 2027 cohort's net earned premium.

The good part, equally real: the default rate fell sequentially for the first time in six quarters (1.17% → 1.16%), the default inventory stabilized (8,044 → 8,020), investment income is growing 21… YoY on a 3.5% book yield, and management repurchased 826,523 shares in Q2 at monthly average prices of … … and … — i.e. at 1.04-1.09x book value, 15… below today's price. Book value per share rose from … to … in 12 months, ….

Estimated value: … – … per share, with the center of gravity at … – … Verdict: MONITOR with a price trigger — not buying at … The company is high quality and impeccably capitalized, but at the current price the median margin of safety is 3%, i.e. zero after any estimation error, and the credit-recession scenario gives …. The threshold at which the thesis becomes actionable with a 20% margin is … — practically the exact zone where management bought in Q2. That's the level to watch, not today's price.

---