Insurance Market Overview — Background for Wedge 2/3 Work

A primer scoped for the financialization thesis. The anchor for the whole document is Preston Wilson at Guy Carpenter / Marsh McLennan, who walked us through the reinsurance stack across May 7 and May 22 and a follow-up May 22 solo session in which he relayed structured input from his firm’s parametric specialist. Where Preston named a structure, an actor, or a piece of math, the rest of this brief is mostly external evidence wrapped around what he surfaced — not the other way around.

1. The trade insurance actually is

The simplest sentence about insurance is the one Preston implicitly assumed and that the financialization primer makes explicit: insurance is risk transfer in exchange for a premium, in exchange for which the insurer holds and invests the money until claims arrive [Synthesis]. Two parties trade: the buyer gives up an uncertain future cost in exchange for a known one today; the seller takes the uncertain cost and is paid both an underwriting fee (premium minus expected losses and expenses) and the right to invest the float between premium collection and claim payout. That is the entire trade. Every line of business, every entity structure, every regulatory regime in §§2-5 is a wrapper around it.

The math is also simple in its shape. A premium $P$ for a one-year policy decomposes into

$$ P = E[L] + E[X] + R + \pi $$

where $E[L]$ is expected loss, $E[X]$ is expected expenses (claims handling, broker commissions, taxes), $R$ is a risk load that compensates the carrier for the variance and tail of the loss distribution, and $\pi$ is the target underwriting profit [Public: Numberanalytics — Ruin Probability primer]. Pure premium itself is generally written as frequency $\times$ severity: $E[L] = \lambda \cdot E[S]$ where claims arrive at intensity $\lambda$ and individual claim severity $S$ has its own distribution.

The first non-trivial question — and one that has to land cleanly for any of the Wedge 2/3 work — is why corporations buy insurance at all. For an individual, risk aversion is a sufficient explanation: people are willing to pay more than $E[L]$ to lock in a certain outcome because their marginal utility of wealth is decreasing. For a public corporation, that argument breaks. Shareholders can already diversify by holding a portfolio of stocks. A widely-held firm should, by Modigliani-Miller, be approximately risk-neutral at the firm level: idiosyncratic shocks wash out in the portfolio. So why does NVIDIA carry $2.8B of warranty reserves, why does TSMC buy property coverage, and why does the Lloyd’s/WTW survey find that 88% of semiconductor risk professionals call supply-chain insurance “mission critical” [Public: Lloyd’s/WTW Loose connections press release, 2023-03]?

The canonical answer is Mayers & Smith’s 1982 Journal of Business paper “On the Corporate Demand for Insurance,” which catalogs the M-M-violating frictions that justify corporate insurance even under diffuse ownership [Public: Mayers & Smith 1982, summarized at SMU Cox]. The list as it applies to the chip industry is:

Bankruptcy and financial-distress costs. Costs of distress scale non-linearly: a 10% drop in asset value at a healthy firm is annoying; the same drop at a leveraged firm can trigger covenant breaches, fire-sale asset dispositions, and dilutive equity raises. Mayers & Smith show that a firm will buy insurance whenever the present-valued reduction in expected distress costs exceeds the loaded premium [Public: Mayers & Smith 1982, SMU synopsis]. The Berk report §6.1 applies this directly to fabs: an outage during a credit downturn can trigger covenant violations and dilutive raises that no shareholder portfolio can diversify away.
Lender requirements. Project finance for a new fab almost always carries insurance covenants. Creditors cannot diversify in the way equity holders can; the bond is a one-way bet on no default. Insurance is therefore demanded by the credit side of the cap table, not the equity side [Synthesis, drawing on Berk §6.1].
Asymmetric information and counterparty pass-through. A fab depending on a sole-source specialty-gas supplier needs to transfer that supplier-failure risk somewhere. Shareholders can’t replicate the pooling. Reinsurers can, because they hold a book of fab-supplier pairs that look statistically independent over the medium term.
Tax and accounting timing. Mayers & Smith point to the convex U.S. corporate tax schedule and the present-value gain from converting volatile pre-tax income into smoother post-tax income; the same logic supports any tool — insurance, derivatives, captives — that smooths reported income [Public: Mayers & Smith 1982, SMU synopsis].
Underinvestment and asset-substitution agency problems. Insurance can mitigate Myers-style underinvestment by removing scenarios in which equity holders walk away from positive-NPV projects because the cash would go to creditors [Public: Mayers & Smith 1982, SMU synopsis].

These are the M-M frictions. The financialization primer §6 makes the operational version of the same point: finance exists to keep capital from sitting idle against a risk that someone else is better-suited to hold. Insurance is the version of that move applied to discrete loss events. Hedging — buying a futures contract, an option, a swap — is the version applied to price exposure. Mathematically they collapse into the same trade: an entity with an exposure pays a premium to lay it off to a counterparty with lower marginal cost of bearing it. The primer puts the unified frame plainly: “every wedge we’re considering — futures, parametric insurance, warranty transfer — is a variation on ‘stop tying up capital against a risk; transfer the risk to someone better suited to hold it.‘”

That unification is load-bearing for the Project TBD thesis because Wedge 2 (warranty transfer) and Wedge 3 (fab/supply-chain insurance) are both versions of this single trade, and the Berk report §8.1 argues that the futures wedge is too. The thing that makes the chip industry different is which frictions dominate: §3 below works through the math, but the punchline is that for the chip downstream the binding frictions are (1) above — distress and rebuild-time costs that exceed available BI indemnity — and (3) — counterparty pass-through where the buyer of a chip is exposed to a supplier’s failure.

One last point in §1 belongs to Buffett, because it shapes the carrier side of the trade in a way that flows through everything else. In Berkshire Hathaway’s 2024 annual letter, Berkshire reports approximately $171 billion of insurance float at year-end 2024, up from $169 billion at year-end 2023 [Public: Berkshire Hathaway 2024 Annual Report]. The 2024 underwriting gain was roughly $9 billion (reinsurance-news cites the underwriting gain as $9bn; the carrier-management piece reports underwriting profit of about $11.4bn including primary lines) [Public: Reinsurance News 2025-02; Carrier Management 2025-03]. The Buffett insight is two-part: (a) underwriting margin is the headline number but (b) the float is the durable asset — money you hold for free, or even cheaper than free if you also earn an underwriting profit, to invest at whatever your portfolio rate is. The financialization primer §7 sketches the same arithmetic on NVIDIA’s warranty book: idle $8B reserve earns nothing; redeployed into next-gen R&D at NVIDIA’s gross margins, the opportunity cost is hundreds of millions a year.

2. Lines of business + biggest players

The global insurance industry collected approximately $7.1 trillion in premium in 2023, with Swiss Re forecasting growth of 2.9% real for life and 3.3% real for non-life into 2024 [Public: Swiss Re sigma 3/2024 World insurance press release; Swiss Re sigma 3/2024 PDF]. Global P&C alone is approximately $2.4 trillion of that total; global life is on track to reach $4.8 trillion by 2035 from roughly $3.1 trillion in 2024 [Public: Swiss Re sigma 3/2024]. Those numbers are the denominator for everything that follows.

The standard cuts are by line of business and by tail length. The economics differ sharply across them, and the cut matters for Wedges 2 and 3:

Life insurance is long-tail and savings-flavored. Premiums are collected over decades; claims (or annuity payouts) are paid out over decades. The product is closer to a structured savings vehicle than to a discrete-loss contract. Investment income on float is the dominant earnings driver.
Health insurance in the U.S. is short-tail (annual or sub-annual claim cycle), regulated heavily under the ACA, and dominated by managed-care economics rather than pure underwriting.
P&C (auto, home, commercial) is the line most relevant to Wedge 3. Auto is short-tail, high-frequency, low-severity. Homeowners is short-tail in claim cycle but with catastrophe-driven severity tails. Commercial property and business interruption — the Preston anchor — is short-to-medium tail with high severity per event.
Specialty (cyber, marine, aviation, political risk, terrorism, professional liability) is heterogeneous but unified by needing custom underwriting per risk. This is where MGAs concentrate because the data and structuring requirements are not yet commoditized.
Reinsurance is the layer above carriers. Reinsurers take on losses ceded by carriers in exchange for a portion of premium. Their advantage is portfolio diversification across geographies, lines, and perils.

Tail length and frequency-vs-severity profile drive how reserves are set, what capital adequacy looks like, and which structures (treaty types, ILS) make sense. Short-tail risks reserve more on chain-ladder; long-tail risks lean on Bornhuetter-Ferguson because chain-ladder is unstable for immature development years [Public: Wikipedia, Bornhuetter-Ferguson] — both methods covered in §3.

Dominant U.S. P&C carriers by direct premium written, 2024:

Rank	Carrier	2024 net/direct premiums written	Notes
1	State Farm	~$107.8B net (mutual)	Auto + home dominant; not stock-listed
2	Berkshire Hathaway (GEICO + Berkshire Hathaway Specialty + Gen Re + National Indemnity)	~$80.4B net; ~$63.3B direct	The Buffett float vehicle; $171B float at YE2024
3	Progressive	~$74.4B net, +21% YoY	Auto-focused; underwriting profit leader
4	Allstate	(top-4 by market share)	Auto + home
5	Travelers	~$41.9B direct, +8.5% YoY	Largest commercial + workers’ comp writer; 8% commercial market share

[Public: Insurance.com top 10 P&C 2026; Carrier Management 2025-03; Insurance Business Magazine]

Top global reinsurers by 2024 gross written premium / reinsurance revenue:

Rank (S&P, 2024 GWP basis)	Reinsurer	2024 GPW (USD)	Notes
1	Swiss Re	$43.1B	Overtook Munich Re on S&P methodology
2	Munich Re	$42.8B	Cited as $51B on alternate methodology
3	Hannover Re	~$36B	German, third-largest reinsurer globally
4-5	Berkshire Hathaway Re / Gen Re	(combined with primary)	Berkshire reinsurance ops part of $171B float
	SCOR	(top-6)	French, asset-intensive
	Lloyd’s of London (market aggregate)	£55.5B GWP (~$70B)	Subscription marketplace, ~$56B capacity 2024
	China Re	top-10	Mainland China dominant
	Tokio Marine	top-10	Japan, expanding internationally

[Public: Reinsurance News on S&P 2024 ranking; Beinsure top-50 reinsurance groups; AM Best world’s largest reinsurers PDF; Lloyd’s Key Facts and Figures]

The ranking accounting-standard caveat matters because Swiss Re and Munich Re report under IFRS 17 and don’t publish a clean GPW figure; S&P imputes one [Public: Reinsurance News]. The point for Wedge 2/3 work is not the rank order but the concentration: roughly half of global reinsurance premium runs through fewer than 10 firms. That concentration is operative for Preston’s “middlemen — and I’m one of them” observation: the value chain is layered (broker → carrier → reinsurance broker → reinsurer → retrocessionnaire → ILS), and the consolidation at each layer is what creates both the relationship-based-trust moat and the disintermediation opportunity.

Specialty insurers relevant to the Wedge 2/3 path:

Firm	Niche	Why it matters here
Beazley	Cyber, marine, political risk	One of three lead U.S. cyber market-shapers (Beazley, Chubb, AXA XL) [Public: SeedPod cyber comparison]
Hiscox	Specialty + ILS	Lloyd’s market specialist; cyber catastrophe consortium with Ariel Re in April 2024 [Public: SeedPod cyber comparison]
Coalition	Cyber MGA	$5B valuation at 2022 Series F led by Allianz X. Aspen Specialty became their U.S. capacity provider in April 2024 [Public: CB Insights on Coalition; Financial Content / Coalition-Aspen capacity agreement, 2024-03]
Munich Re (specialty side)	Battery warranty + parametric	Reinsured Hithium’s 15-year ESS performance warranty in 2023; TWAICE data partnership [Public: Munich Re + Hithium press, 2023-10-26; TWAICE Munich Re partnership]

Coalition is the closest existing precedent for the MGA structure Preston pointed at in May 22 and which the Berk report §6.4 proposes for both fab and warranty wedges. The MGA earns a fee, prices the policy using a proprietary data layer (claim history, network telemetry), and writes on a carrier’s or reinsurer’s balance sheet — keeping its own capital footprint small and avoiding the full insurance company regulatory perimeter.

Lloyd’s specifically is worth lingering on, because the May 22 solo session hinted at it as a market structure the Wedge 3 product might one day need. Lloyd’s is not a single insurer; it is a subscription marketplace of syndicates (capital pools managed by managing agents) that each take a slice of any given risk via a “slip” that a broker walks around the underwriting room. Lloyd’s reported £55.5 billion of gross written premium in 2024 (about $70B at recent FX), pre-tax profit of £9.6 billion, and total market capacity of £56 billion [Public: Lloyd’s Key Facts and Figures; Insurance Times new syndicate report; LMA Lloyd’s 2025 Insights Report]. The top-10 syndicates’ share of capacity fell from 39% in 2023 to 37% in 2024, signaling capacity moving to smaller specialty syndicates [Public: Insurance Times]. For a novel risk class — say, parametric fab equipment-failure coverage — Lloyd’s is the venue most likely to entertain it before standard markets will.

3. The math

The math reduces, mostly, to four buckets: ratemaking (set $P$ so the trade is profitable), reserving (estimate what you owe on claims already incurred), capital adequacy (hold enough surplus to survive bad years), and treaty arithmetic (how layers of cession share the loss). Preston’s framework in May 22 — the four-pillar parametric test — sits on top of all four.

3.1 Ratemaking

The pure premium is

$$ P_{\text{pure}} = \lambda \cdot E[S] $$

where claim frequency $\lambda$ is the expected number of claims per exposure unit and $S$ is severity. The gross premium adds expenses, risk load, and target profit:

$$ P_{\text{gross}} = \frac{P_{\text{pure}} + \text{fixed expenses}}{1 - v - \pi} $$

where $v$ is variable expense ratio and $\pi$ is target profit margin. The variance of the loss distribution shows up in two places: in the risk load $R$ added to pure premium for any single contract, and in the required capital (§3.3).

When an actuary has some loss experience for a specific risk but not enough to fully trust it, the standard tool is credibility weighting: blend the observed experience with a broader manual rate.

$$ P_{\text{credible}} = Z \cdot P_{\text{experience}} + (1-Z) \cdot P_{\text{manual}} $$

The credibility factor $Z \in [0,1]$ rises with the volume of exposure and the homogeneity of the book. For something like fab equipment-failure parametric coverage, $Z$ would start at zero — there is no historical loss series — and that is what Preston’s specialist meant in May 22 when she said “no model” was one of the three missing pillars. The “model” pillar is exactly the actuarial model needed to translate sensor readings into expected payouts that a reinsurer will write against.

3.2 Loss ratio and combined ratio

Two ratios run the carrier P&L.

$$ \text{Loss ratio} = \frac{\text{incurred losses + LAE}}{\text{earned premium}} \qquad \text{Expense ratio} = \frac{\text{underwriting expenses}}{\text{written premium}} $$

$$ \text{Combined ratio} = \text{Loss ratio} + \text{Expense ratio} $$

A combined ratio below 100% means underwriting was profitable. Above 100% means the underwriting lost money, but the carrier may still be profitable from investment income on float — exactly the Berkshire structure described in §1. Berkshire’s $171B float earning a portfolio return is the headline; the $9-11B underwriting gain on top is the bonus.

3.3 Reserving — chain-ladder and Bornhuetter-Ferguson

Most claims aren’t paid the day they occur. Reserving estimates the ultimate cost of claims already incurred (whether reported or not — IBNR, “incurred but not reported”). The two canonical methods:

Chain-ladder. Build a triangle of cumulative paid (or incurred) losses by accident year and development year. Compute age-to-age development factors $f_j$ as the ratio of cumulative losses at development year $j+1$ to development year $j$. Project ultimate by multiplying current observed losses by the product of remaining $f_j$ [Public: Wikipedia, Loss reserving; Fiveable, chain ladder + BF study guide]. Chain-ladder works well for mature accident years with stable development; it fails for immature years where one or two large early claims can distort the ratios.

Bornhuetter-Ferguson. Blend chain-ladder with an a priori expected loss ratio. Estimate ultimate as

$$ \text{Ultimate} = \text{Reported losses} + (\text{a priori expected losses}) \cdot (\text{percent unreported}) $$

The “percent unreported” comes from the chain-ladder development pattern; the a priori loss ratio comes from external benchmarks. BF is the standard for new lines and immature accident years for exactly the reason chain-ladder fails on them: it anchors the unreported piece to an external expectation rather than leveraging tiny early claim counts [Public: Wikipedia, Bornhuetter-Ferguson; Sigma Actuarial, BF method].

This matters for Wedge 2. NVIDIA’s warranty reserve has grown 7× in two years per the Berk §5.2 numbers from the FY2026 10-K. A reinsurer pricing a warranty risk-transfer deal on a young book like this would be on Bornhuetter-Ferguson territory, not chain-ladder.

3.4 Ruin theory — Cramér-Lundberg

How much capital does an insurer need to survive bad years? The classical model is Cramér-Lundberg [Public: Wikipedia, Ruin theory; Numberanalytics primer]. The surplus process is

$$ U(t) = u + c \cdot t - \sum_{i=1}^{N(t)} S_i $$

where $u$ is initial surplus, $c$ is premium income rate, $N(t)$ is a Poisson process of claim arrivals with intensity $\lambda$, and the $S_i$ are i.i.d. claim severities. Ruin is the first time $U(t) < 0$. Lundberg’s inequality bounds the ruin probability:

$$ \psi(u) \leq e^{-Ru} $$

where the adjustment coefficient $R$ is the positive root of $\lambda \cdot M_S(r) = \lambda + c \cdot r$ and $M_S$ is the moment generating function of severity. The intuition: bigger initial surplus, smaller ruin probability, exponentially. The practical implication: modern regulatory capital (RBC, Solvency II SCR) is the descendant of this model, dressed up with multiple risk modules.

3.5 Capital adequacy — RBC and Solvency II

The U.S. NAIC adopted Risk-Based Capital (RBC) in 1993, calculated by summing capital charges across asset risk, underwriting risk, credit risk, and (for some lines) business risk, then aggregating via a covariance adjustment that recognizes diversification [Public: NAIC RBC overview; NAIC RBC Preamble]. The basic RBC formula for a P&C insurer aggregates as

$$ \text{RBC} = R_0 + \sqrt{R_1^2 + R_2^2 + R_3^2 + R_4^2 + R_5^2} $$

where the $R_i$ are charges for affiliated investments, fixed-income, equity, credit/reinsurance, reserves, and net written premium respectively [Public: NAIC RBC Preamble]. The square-root aggregation is the diversification benefit.

Solvency II in the EU is the more rigorous descendant. The Solvency Capital Requirement (SCR) is calibrated to a 99.5% one-year Value-at-Risk (VaR):

$$ \text{SCR} = \text{VaR}_{0.995}(\text{Basic Own Funds, 1-year horizon}) $$

That is, hold enough capital that you survive the next year with probability 99.5% — a “once-in-200-years” calibration [Public: EIOPA SCR overview; Skadden Solvency II guide chapter 8]. The Standard Formula aggregates modular risk capital across market, counterparty default, life, health, non-life, and operational risk modules, using EIOPA-provided correlation matrices. Insurers may instead use an internal model approved by the supervisor; large reinsurers (Swiss Re, Munich Re, SCOR) typically do.

The asset side of the Cramér-Lundberg intuition is now: capital regulators set $u$ (initial surplus) such that the ruin probability is below regulator tolerance, given the firm’s actual $\lambda$ and severity distribution.

Tail Value-at-Risk (TVaR), defined as

$$ \text{TVaR}\alpha = E[L \mid L > \text{VaR}\alpha] $$

is increasingly used as a coherent alternative to VaR. The Swiss Solvency Test, for instance, uses TVaR at 99% rather than VaR at 99.5%. Coherent in the Artzner-Delbaen-Eber-Heath sense means it satisfies subadditivity, which VaR doesn’t always.

3.6 Reinsurance treaty math

Reinsurance comes in two big buckets: proportional (the reinsurer shares premium and loss by a defined percentage) and non-proportional (the reinsurer pays losses above an attachment) [Public: Accelerant, types of reinsurance; Risk Coverage Hub, treaty structures].

Treaty type	Mechanics	Math	When used
Quota share (proportional)	Reinsurer takes fixed $q%$ of premium and loss on every risk	$L_{re} = q \cdot L$, $P_{re} = q \cdot P$	New lines (cedent wants surplus relief); growing portfolios; risks with stable expected loss
Surplus share (proportional)	Cedent retains a “line” $\ell$ per risk; reinsurer takes proportional excess up to defined multiple of $\ell$	If TIV $= V$, $L_{re} = L \cdot \max(0, V-\ell)/V$ (subject to multiple-of-line cap)	When risks vary widely in size; cedent wants per-risk balance
Excess-of-loss (non-proportional)	Reinsurer pays losses above retention $R$ up to limit $M$, per risk or per occurrence	$L_{re} = \min(\max(L-R, 0), M)$	Catastrophe protection; high-severity, low-frequency risks
Stop-loss (non-proportional)	Reinsurer pays aggregate annual loss above a defined ratio or amount	$L_{re} = \min(\max(\sum L - A, 0), M)$	Hedging total portfolio loss; small portfolios; agricultural / health

Preston’s May 7 stack diagram runs from cedent through carrier through reinsurer through retrocessionnaire to ILS investors — each layer pricing its own slice of the loss distribution. The ILS bond is structurally an excess-of-loss layer with collateralized capacity sourced from capital markets rather than a reinsurer’s balance sheet.

The four-pillar test from May 22 sits on top of all of this: even before treaty math kicks in, a parametric trigger has to clear (1) a measurable metric, (2) a trusted observer, (3) a loss model, and (4) reinsurer market acceptance — i.e. a credible $P_{\text{credible}}$ that an actuarial team will sign off on.

4. Ways to structure an insurance business

4.1 Carrier organizational form

Form	What it is	Examples relevant here
Stock insurer	Owned by shareholders; profits to equity	Berkshire Hathaway, Allstate, Progressive, Travelers, AIG
Mutual insurer	Owned by policyholders; profits returned via dividends or rate reductions	State Farm, Liberty Mutual, MassMutual (life)
Reciprocal	Subscribers exchange contracts of indemnity; managed by an attorney-in-fact	USAA (historically), Farmers Insurance Exchange
Lloyd’s syndicate	Capital pool at Lloyd’s marketplace managed by a managing agent, writing through the Lloyd’s chain of security	Beazley Syndicate 623; Hiscox Syndicate 33
P&I club	Mutual marine protection-and-indemnity associations; ship-owner mutuals	Standard Club, North P&I, Gard

The mutual form was historically dominant in U.S. P&C because it aligned policyholder and owner. Stock form dominates global commercial and reinsurance because it can raise external capital. The Lloyd’s syndicate is unique — Names and corporate members commit capital to a one-year-of-account syndicate, which writes business through the marketplace and reinsures across syndicates [Public: Lloyd’s, Syndicate page; Lloyd’s, How the market works]. The subscription model that Lloyd’s pioneered — a broker walking a slip around the room, each syndicate taking a line — is the structural template for ILS placements today.

4.2 The carrier / MGA / broker stack

Role	What they do	Capital	Regulatory burden
Carrier (insurance company)	Issues policies, takes balance-sheet risk	Heavy; subject to RBC/Solvency II	Full state license + Schedule P/SCR
Reinsurer	Takes risk from carriers	Heavy; same regime, often Bermuda/Solvency II	Same as carrier, plus credit-for-reinsurance compliance
MGA (Managing General Agent)	Underwrites and binds policies on behalf of a carrier	Light (fee-based, no balance sheet)	Producer license + carrier oversight
MGU (Managing General Underwriter)	Sub-type of MGA, typically specialty with underwriting authority on more complex risks	Light	Similar to MGA
Fronting carrier	Licensed carrier that issues paper but cedes most/all risk to a reinsurer behind it	Heavy on paper, light in practice	Full carrier license; fronting fees typically 4-8%
Broker (retail / wholesale / reinsurance)	Intermediates between buyer and carrier; reinsurance brokers between carrier and reinsurer	None	Producer license

Preston’s value-chain map in May 7 — original insured → retail broker → carrier → reinsurance broker → reinsurer → retrocessionnaire → capital markets — assigns each role above to a distinct firm. Guy Carpenter, where Preston works, is a reinsurance broker. Marsh McLennan, its parent, is the world’s largest broker group at all three layers. The MGA path Preston repeatedly came back to — “if you have a good model that no one else has… the best place for you to sit would be to be the market, to be the reinsurer writing the parametric” May 22 solo — is the path that minimizes regulatory burden while maximizing economic exposure to the underwriting result.

4.3 Captive insurance

A captive is an insurance company a corporation (or group) sets up to insure its own risks [Public: Carey Olsen, Bermuda captive insurers; IFC Review, Cayman captive]. Three main flavors:

Single-parent captive. Insures only the parent and group entities. Used by large multinationals to retain risk, access reinsurance at wholesale rates, and gain tax benefits.
Group captive. Multiple unrelated companies share a captive; common in industries with similar risk profiles (trucking, healthcare).
Cell captive / segregated cell / protected cell. A core company holds many legally-separated cells, each with its own assets and liabilities. Bermuda calls these “segregated account companies”; Cayman calls them “segregated portfolio companies”; an incorporated cell variant in Cayman is the “portfolio insurance company” [Public: Wikipedia, Cell captive; Loeb Smith, Cayman captive briefing].

The Berk §6.2 flags captives as the “default for large fab operators self-insuring tail risk.” Hyperscalers and IDMs reportedly do this. For TBD’s purposes the captive isn’t a startup play directly; it’s the competition a new insurance product has to beat on price and coverage scope.

4.4 ILS — cat bonds, sidecars, ILWs, collateralized re

Insurance-linked securities are the route from insurance risk to capital markets directly, bypassing some of the reinsurance layers Preston critiqued in May 7. The structures:

Cat bonds. A special-purpose insurer (SPI) issues a bond. Investors pay principal; the SPI invests it in collateral. If a defined catastrophe occurs (parametric, industry-loss, or indemnity trigger), the SPI uses the collateral to pay the cedent and bondholders lose part or all of principal. If not, bondholders receive periodic coupons plus principal at maturity. Outstanding cat bond market was $49.5 billion at end of 2024, after a record $17.7 billion of new issuance in 2024 alone [Public: Artemis, 2024 cat bond record; Artemis, $17.7B issuance].
Sidecars. Off-balance-sheet vehicles that reinsure a portion of a sponsor reinsurer’s book on a proportional or quota-share basis. Investors provide capital for a single underwriting year. Sidecar market grew ~70% to ~$17 billion in mid-2024 per Aon Securities; AM Best put 2024 sidecar capacity at $6-8 billion (the discrepancy reflects different counting methodologies — gross capacity vs. net capital at risk) [Public: Artemis, sidecar 70% growth; Artemis, ILS capacity recovery].
Collateralized reinsurance. Reinsurance where the reinsurer fully collateralizes its obligations in a segregated trust. ILS funds use this structure to write reinsurance directly without becoming a regulated reinsurer themselves.
Industry loss warranties (ILWs). A parametric-style derivative paying when an industry loss index crosses a threshold. Example: “$100M limit US wind ILW attaching at $20B” pays if total industry losses from a single hurricane exceed $20B [Public: Wikipedia, ILW; Artemis library, ILW]. ILWs often have dual triggers — an industry index and a policyholder-specific loss — to balance basis risk against moral hazard.

4.5 Parametric vs indemnity

The structural distinction Preston centered on in both May 22 conversations:

Indemnity coverage. Insurer pays actual assessed loss. Requires claims adjustment, loss documentation, often litigation. The financialization primer §8 notes the standard claims cycle is “slow, requires claims adjustment and proof of damage.”
Parametric coverage. Insurer pays a pre-agreed amount the instant a measured parameter crosses a threshold. No adjuster. Payout in days, not months. Basis risk is the trade-off: trigger fires without loss (positive basis risk) or loss without trigger (negative basis risk). Positive basis risk is often written out of policy text on good-faith grounds Preston May 22.

The four-pillar parametric framework — metric, measuring agent, model, market — is the operational test. Preston’s specialist named this in May 22: every parametric application must clear all four. For nat-cat risks (earthquake, windstorm, tropical cyclone) all four exist. For man-made equipment failure (fab overheating, GPU thermal events), three of four are missing.

4.6 Reinsurance treaty vs facultative

One last cut, because Preston works in facultative specifically. A treaty is an automatic, multi-policy agreement: the carrier cedes a defined portion of an entire book to a reinsurer, who has no per-risk veto. A facultative reinsurance is single-risk: each individual policy is presented to the reinsurer, who decides yes/no/at-what-price. Facultative is used for large, complex, or unusual risks where the reinsurer wants to underwrite each one — large commercial property programs being the canonical example, which is why Preston does it for a living. New product categories (semiconductor fab parametric, GPU warranty) would almost certainly start facultative before they become large or stable enough for treaty placement.

5. Regulation

5.1 U.S. — state primacy, federal overlay

The dominant U.S. fact is McCarran-Ferguson (1945), which gave states primary authority to regulate insurance, codified federal deference except where Congress specifically intervenes. The NAIC is the state regulators’ coordinating body; it writes Model Laws and Regulations that states adopt with variation [Public: NAIC RBC overview; NAIC Surplus Lines]. Federal overlay arrived with Dodd-Frank (2010), which created the Federal Insurance Office (FIO) inside Treasury — not a regulator, but a monitor with international representation and a limited preemption authority for international agreements.

The federal/state line has practical implications:

Admitted vs surplus lines. Admitted insurers are licensed in each state where they write business and subject to state rate and form approval. Surplus lines insurers are not state-licensed but are accessed via state-licensed surplus lines brokers for risks the admitted market won’t cover. Surplus lines lets specialty product flow without rate/form approval friction — most fab and supply-chain insurance for novel risks routes through surplus lines [Public: NAIC Surplus Lines; Surplus Manual 2026; AgentSync surplus lines compliance].
Schedule P. Annual statutory loss-reserve schedule by line of business and accident year that P&C insurers file with state regulators. Public via NAIC; the single most important data artifact for any actuarial benchmarking. A Wedge-3 MGA underwriting on top of a fronting carrier would have its book reflected in that carrier’s Schedule P, which is part of why fronting-carrier selection matters.
RBC (§3.5 above). State-administered, NAIC formulaic.

5.2 Europe — Solvency II, EIOPA

Solvency II (Directive 2009/138/EC, in force since 2016, recently revised) is the EU’s risk-based regulatory regime [Public: EIOPA SCR overview; Skadden Solvency II guide]. Three-pillar structure:

Quantitative requirements (Pillar 1): the SCR and MCR (Minimum Capital Requirement) we discussed in §3.5.
Governance and risk management (Pillar 2): ORSA (Own Risk and Solvency Assessment).
Disclosure and supervisory reporting (Pillar 3): SFCR public reports.

EIOPA is the pan-EU supervisor coordinating national authorities (BaFin in Germany, ACPR in France, etc.).

5.3 UK — PRA, FCA, Lloyd’s Acts

Post-Brexit the UK runs its own regime, broadly equivalent to Solvency II but increasingly diverging. The Prudential Regulation Authority (PRA), part of the Bank of England, sets capital and solvency standards. The Financial Conduct Authority (FCA) handles conduct. Lloyd’s of London has its own statutory framework via successive Lloyd’s Acts; the PRA supervises the Lloyd’s market alongside the Council of Lloyd’s [Public: Lloyd’s market overview].

5.4 Bermuda — BMA, equivalence, reciprocal jurisdiction

Bermuda is the dominant offshore domicile for both commercial reinsurance and ILS structures. The Bermuda Monetary Authority (BMA) regulates Class 4 (large commercial reinsurer) and Special Purpose Insurer / Collateralized Insurer classes; the Class 3A and 3B captive classes are also large. Bermuda achieved Solvency II equivalence for its commercial classes (not captive/SPI) [Public: Carey Olsen Bermuda captives; Skadden BMA papers, 2024-10; Mayer Brown on asset-intensive reinsurance, 2026-03].

In 2019 the NAIC updated its Credit for Reinsurance Model Law to recognize Reciprocal Jurisdictions — including Bermuda and the EU — under which qualifying reinsurers from those jurisdictions don’t need to post collateral to U.S. cedents [Public: NAIC Bermuda Reciprocal Jurisdiction Findings; Conyers, NAIC Reciprocal Jurisdiction]. Before this, non-U.S. reinsurers had to fund a U.S. trust or letter-of-credit equal to their gross liabilities; reciprocal jurisdiction status drops that to zero in the qualifying classes. This change roughly halved the capital friction of running a Bermuda reinsurer over a U.S. cedent base.

The U.S.-EU Covered Agreement (and a separate U.S.-UK version) operates on parallel logic with a federal preemption hook — states had 60 months from signing to eliminate reinsurance collateral or face FIO preemption [Public: Mayer Brown 2026-03].

5.5 CFTC vs insurance regulator — the boundary for ILS and parametric

This is the regulatory question most directly relevant to Wedge 3. A parametric product might be insurance or might be a derivative. The legal test is essentially: does the policyholder need to demonstrate an insurable interest and a proof of loss? If yes, it’s insurance, state-regulated. If no — if the contract just pays on a triggering event regardless of the holder’s actual loss — it’s a derivative, CFTC-regulated under the Commodity Exchange Act [Public: Schupp, Medium on parametric regulation; NatLawReview on weather risk regulation; Wilmer Hale, CFTC event contracts].

The practical implication for any Wedge 3 product: the trigger design choice (parametric vs hybrid vs indemnity) determines which regulator you face. A pure parametric — “if temperature crosses X for Y minutes, $100M pays” — without an insurable-interest hook is at risk of being a “swap” under CFTC jurisdiction. A parametric plus an insurable-interest requirement (the policyholder must own the fab and certify a covered event has occurred) stays in the insurance camp. Most commercial parametric structures today are written with the insurable-interest hook precisely to stay within state insurance regulation.

5.6 TRIA, OFAC

TRIA — the Terrorism Risk Insurance Act — provides a federal backstop for commercial property catastrophic terrorism losses, currently authorized through Dec 31, 2027 [Public: Congressional Research Service, TRIA primer; GAO 2024 report on TRIA reauthorization; NAIC TRIA topic page]. Treasury reported in 2024 that 74% of stand-alone terrorism policies were TRIP-eligible. For any commercial property line touching the U.S., TRIA is a structural fact; new fab-focused products would slot in alongside it.

OFAC runs the U.S. sanctions regime. Insurance contracts with sanctioned counterparties are prohibited or require specific licenses. This is also where the historic compliance wedge fed into insurance — knowing the end-user of a chip matters under both OFAC and EAR. The 2026-05 compliance wedge killdown doesn’t change the regulatory fact: any insurance product written by a U.S.-domiciled entity has to OFAC-screen counterparties.

6. Where the evidence is thin and what disconfirms the thesis

The point of an insurance-market primer for Wedges 2 and 3 isn’t to confirm them. It’s to let the founders read external evidence faster when the next interview surfaces a structure or a term. Three reminders before that work continues:

Disconfirming voice 1 — Jeremy Jawish on parametric demand. In May 22 Jawish — co-founder/CEO of Shift Technology, a fraud-detection MGA-adjacent — said directly: “The parametric market is still small and it’s not worth it. People are just not comfortable with parametric triggers for insurance policies.” He added that claims processing is only 15% of premium cost, meaning innovations in claim efficiency have a limited addressable base. He noted even Coalition’s parametric cyber market is small. He was relaxed about it, not adversarial — flat affect of someone who has already investigated and closed the door per the debrief. This is the strongest disconfirming voice in the entire vault on Wedge 3’s parametric framing, and it sits in tension with the Lloyd’s/WTW 88%-mission-critical demand signal. Both can be true: the need is real and the current market is small. But the gap between need and adoption is exactly the buyer-side conservatism Jawish flagged — they want best price over simplicity — and that maps onto a real product-design constraint.

Disconfirming voice 2 — Preston’s own soft-market caveat. Preston himself flagged in May 7 that commercial property rates are down 25-30% over the past 2-3 years. Carriers are chasing premium in data centers and semiconductor supply chain. Any new product entering this market does not enter into pricing vacuum — it enters into an actively softening market where incumbents are already cutting rates. A differentiated structure that’s slightly more expensive than the cheapest indemnity quote may struggle to win share until the cycle turns.

Disconfirming voice 3 — the gap that has no voice. We have zero demand-side interviews. Preston is a sell-side intermediary. The May 22 debrief makes this explicit: “Preston is an insurance intermediary, not a semiconductor fab operator or procurement officer. His enthusiasm for the ILS/parametric structure is a supply-side signal… but does not yet constitute demand-side validation from the actual buyer of insurance at a fab.” That gap is the single biggest hole in the brief. Until a fab CFO or risk manager prices a real parametric quote against a real indemnity quote and tells us which one they’d buy, the entire Wedge 3 willingness-to-pay rests on the Lloyd’s/WTW survey and Preston’s sell-side enthusiasm.

Where this brief is softest. A few specific places where I’d want the founders to read with extra skepticism:

The Mayers-Smith friction taxonomy is theoretically clean but lacks Project-TBD-specific empirical evidence on which friction is binding for chip buyers. Section 1 lists the canonical M-M frictions but doesn’t cite a single chip-industry buyer who explicitly named them.
The size-of-market figures for parametric insurance vary 2-3× across analyst houses ($14B - $20B as of 2024). I cited multiple ranges; treat them as the same order-of-magnitude estimate rather than precise data.
The Bermuda/Solvency II equivalence discussion is the cleanest piece of regulation but also the most likely to shift; reciprocal jurisdiction and covered-agreement implementation continues to evolve. Verify with current counsel before any structuring step.
The CFTC/insurance-regulator boundary on parametric is genuinely unsettled. There is no bright-line test, only facts-and-circumstances guidance. A founder building a parametric product should consult both a securities lawyer and an insurance regulatory lawyer in any structuring conversation.

7. Open questions to keep on the list

Per RDI discipline this brief ends with questions, not conclusions.

Buyer-side WTP, not supply-side. Who, specifically, would we want to talk to at a fab CFO/risk-management level to validate the gap between the Lloyd’s/WTW survey result (88% mission-critical) and Jawish’s market-still-small observation? Preston offered intros in May 7 — are those intros to brokers/clients of his, or to internal Marsh McLennan parametric specialists?
The float math for Wedge 2. Berkshire holds $171B float at a portfolio rate of (call it) 5-6%; the underwriting is approximately break-even-to-modestly-profitable on average. What’s the equivalent arithmetic for a hypothetical NVIDIA-warranty specialist? Assuming the Berk §5.4.2 illustration of $2.5B/year accrual transferred, the float would grow at $2.5B/year with claims paying out over a 2-3 year tail — what target portfolio return makes the structure viable, and what reinsurance backing brings the risk load to a level a 3-year-tail underwriter can hold?
Reinsurance capacity for a novel line. Lloyd’s syndicate capacity ran £56B in 2024 with small/mid syndicates growing share. Is the right first-write entry for a novel semiconductor parametric product (a) a Lloyd’s syndicate willing to write a tiny initial line, (b) a Bermuda special-purpose insurer / collateralized re structure backed by an ILS investor, or (c) a fronting carrier with a reinsurance backstop? Preston implied (b) in May 7 but the May 22 solo suggested the MGA-into-reinsurer path. These are not the same answer.
Coalition’s pure-parametric share. Jawish flagged Coalition’s parametric cyber book as “small.” Public data on the Coalition-Aspen capacity arrangement names neither the parametric share nor the loss ratio. If Coalition is the structural template, what does their actual product mix look like today? This is a fact we can probably surface via secondary sources.
The “be the measurement agent” path. Preston explicitly raised the moral hazard problem in May 22: you cannot be measurement agent, modeler, and insurer. The MGA structure resolves this by keeping the measurement layer organizationally separate. PDF Solutions surfaced in Berk §6.4 as the natural fab measurement candidate. What does a real commercial relationship with PDF Solutions look like — licensing, JV, equity-linked partnership? This is concrete enough to be worth asking PDF directly.

8. Confidence summary

The shape of the global insurance market in §§1-5 is well-supported by external evidence. Premium figures, reinsurer rankings, ILS market sizes, regulatory frameworks are all multiply-sourced. The math in §3 is textbook material with multiple authoritative sources.

The link from this market structure to the Project TBD Wedges 2 and 3 is supported by exactly one practitioner (Preston Wilson, three sessions) plus indirect signals from Berk-report sources. Preston is high-quality sell-side input but he is not a buyer, an actuary, or a regulator. Until a fab-side or NVIDIA-side principal validates a price quote, the Wedge 3 willingness-to-pay rests on the Lloyd’s/WTW survey alone, and the Wedge 2 willingness-to-pay rests on NVIDIA’s reserve trajectory plus the Munich Re/Hithium structural analogue.

The two pieces of disconfirming evidence in §6 — Jawish’s parametric skepticism and Preston’s soft-market caveat — are not fatal but they discipline the brief: Wedge 3 is structurally well-founded but commercially still early. Wedge 2 is commercially better-signaled (NVIDIA reserve, Munich Re/Hithium) but structurally requires the MGA + reinsurer + measurement-agent triangle to all be built, and we have zero direct buyer-side validation that NVIDIA would actually transact at the prices a specialist could write at.

Sources

Primary internal interviews

Marsh McLennan), 2026-05-07 — reinsurance stack; ILS + parametric proposal; soft-market dynamics
Preston Wilson, 2026-05-22 — risk buckets (property, BI, CBI); moral hazard; market sizing
Preston Wilson, 2026-05-22 solo — four-pillar parametric test; basis-risk taxonomy; three positioning paths
Jeremy Jawish (Shift Technology), 2026-05-22 — disconfirming voice on parametric

Internal synthesis and primer

Berk independent study report, 2026-06-09 — §5 (Wedge 2), §6 (Wedge 3)
Financialization primer, 2026-05-29 — M-M frictions, float math, parametric basics
Max Mirgoli briefing, 2026-05-15 — Munich Re / TWAICE / Hithium framing
Reverse logistics warranty TAM, 2026-05-29 — Wedge 2 TAM

External

Sources: Preston Wilson May 7; Preston Wilson May 22; Preston Wilson May 22 solo; Jeremy Jawish May 22; Berk independent study report; financialization primer; Max Mirgoli briefing

Project TBD Memory Vault

Explorer

Insurance Market Overview — Background for Wedge 2/3 Work

Insurance Market Overview — Background for Wedge 2/3 Work

1. The trade insurance actually is

2. Lines of business + biggest players

3. The math

3.1 Ratemaking

3.2 Loss ratio and combined ratio

3.3 Reserving — chain-ladder and Bornhuetter-Ferguson

3.4 Ruin theory — Cramér-Lundberg

3.5 Capital adequacy — RBC and Solvency II

3.6 Reinsurance treaty math

4. Ways to structure an insurance business

4.1 Carrier organizational form

4.2 The carrier / MGA / broker stack

4.3 Captive insurance

4.4 ILS — cat bonds, sidecars, ILWs, collateralized re

4.5 Parametric vs indemnity

4.6 Reinsurance treaty vs facultative

5. Regulation

5.1 U.S. — state primacy, federal overlay

5.2 Europe — Solvency II, EIOPA

5.3 UK — PRA, FCA, Lloyd’s Acts

5.4 Bermuda — BMA, equivalence, reciprocal jurisdiction

5.5 CFTC vs insurance regulator — the boundary for ILS and parametric

5.6 TRIA, OFAC

6. Where the evidence is thin and what disconfirms the thesis

7. Open questions to keep on the list

8. Confidence summary

Sources

Primary internal interviews

Internal synthesis and primer

External

Graph View

Table of Contents

Backlinks