If you sit on the board of a group captive or a risk retention group, or if you manage one as an outside captive manager, you have been handed a reserve number labeled IBNR and asked to sign off on it. This article explains what that number is for a pooled structure and what to expect from the annual reserve study.
A group captive is a licensed captive owned by multiple unrelated companies, usually in the same industry (trucking, construction, healthcare), that pool risk inside one vehicle. A risk retention group (RRG) is a federal structure created by the Liability Risk Retention Act of 1986, liability only, owned by its members. Both differ from a single-parent captive, covered separately. IBNR is the sum of claims incurred but not yet reported plus expected development on known claims, and actuaries use that broad definition almost universally (Friedland, p. 388). If the letters are new, the plain-English IBNR explainer is the right starting point.
Why pooling changes the reserving problem
A group captive’s single loss triangle contains the claim experience of every member. That sounds like a computational convenience, and it is, but it hides the editorial tension that runs through every reserve report in this market. The actuary is reserving for the pool as a whole, because that is what the domicile regulator, the auditor, and the captive’s balance sheet require. The board is thinking about individual members, because most group captives return surplus to members through dividends or experience-rating mechanisms, and a member who thinks they are subsidizing another member’s losses will leave. A good actuary produces both views and reconciles them. A mediocre one delivers the aggregate number, lets the member allocation happen somewhere off to the side in a separate spreadsheet, and does not explain how the two views line up.
That tension is the single most important thing for a captive manager or member board to internalize. Aggregate reserves answer the auditor’s question: what does the captive owe at the valuation date? Member-level experience answers the board’s question: is the captive fair to every member, is the dividend formula defensible, and is anyone being asked to pay for losses they did not cause? Both questions are legitimate. Neither is optional. If your reserve study gives you only the aggregate side, you are missing the governance conversation that is arguably the reason your group exists in the first place.
The CAS Forum paper on ratemaking for captives and alternative market vehicles covers the pricing side of this tension, and many of the same techniques translate directly to reserving.
Data credibility sits between a single-parent captive and a real insurer
Pooling improves credibility relative to a single-parent captive because the claim count is larger. A ten-member group captive writing commercial auto liability will see more claims in a year than any one member did on its own, which makes the development patterns in the triangle less noisy and the tail selection less of a judgment call. This is a real advantage and it is one of the reasons group captives exist in the first place.
It does not mean the data is as rich as a full insurer’s book. The mix is still heterogeneous compared with an admitted carrier that has written the same line for decades. And group captives are not static: members enter, members leave, and the pool that exists today is usually not the pool that generated the oldest accident years in the triangle. Friedland’s warning about thin data pushing toward Bornhuetter-Ferguson and expected claims on recent years and low-frequency lines (Friedland, p. 152) applies directly, even though the pool is larger than any of its members would be alone.
The practical implication is that the actuary will still lean on Bornhuetter-Ferguson and expected claims for the most recent accident years, and on chain ladder for the mature years. The pool is larger, but the same credibility instincts apply. If your reserve report uses chain ladder on the latest accident year for a long-tail line, that is a yellow flag worth a question.
The methods and a pooling-specific wrinkle
Group captive reserving uses the same core toolkit as any other property and casualty reserving engagement. Chain ladder is the workhorse for mature accident years, built on the assumption that claims already recorded will continue to develop in the same pattern (Friedland, p. 84). Bornhuetter-Ferguson is the workhorse for recent and intermediate years, anchoring the unreported portion to an expected-claims estimate rather than to a leveraged development factor (Friedland, p. 152). Expected claims is the right answer for a brand-new line or a brand-new cohort of members where there is not yet credible data (Friedland, p. 131). Friedland states directly that no single method can produce the best estimate in all situations (Friedland, p. 345), and a competent reserve study shows you which method was selected for each accident year and why. If you want a plain-English walk through how the triangle itself behaves under chain ladder, the triangle explainer is the best starting point.
The pooling-specific wrinkle is that when membership has changed materially over the experience period, the historical triangle reflects a different pool than the current one. A captive that had eight members in 2018 and now has twelve, three of them new since 2024, does not have a clean ten-year history of the current book. Friedland’s general principle on a changing mix of business applies directly: any change in the composition of the underlying risk invalidates the chain ladder assumption that past development patterns project the future. The fix is not mysterious. The actuary either reweights historical data to approximate the current mix, applies expected claims to the current member mix for recent years, or segments the triangle by membership cohort. Whichever approach is chosen should be documented in writing. If the reserve report treats ten years of history as though it were a single pool, ask how membership change was handled.
The fronting and reinsurance structure, and why it dominates the estimate
This is the single most important technical section for a captive manager or member board to understand, because it explains why the captive’s reserves are not the same as the gross losses of the underlying business.
Most group captives do not issue policies directly. They front through an admitted commercial carrier, which issues the policies to each member and then reinsures the risk back to the captive. The fronting carrier takes a fee, often retains some slice of the risk, and provides the admitted paper that the members’ insurance certificates reference. The captive assumes what actuaries call the “working layer,” the slice of the loss distribution where claims actually happen at meaningful frequency. Above the working layer, most group captives buy per-occurrence excess-of-loss reinsurance with a retention at some level such as 500,000 or one million dollars. And above all of that, many group captives also buy an aggregate stop-loss that caps the captive’s total annual liability at a multiple of expected losses.
That three-layer structure is exactly the one Friedland addresses when she describes the reserving pattern for a self-insured pool with a combination of per-occurrence excess-of-loss and aggregate stop-loss coverage. The standard approach is to estimate ultimate claims limited to the per-occurrence retention first, then apply the stop-loss limit as a final adjustment (Friedland, p. 332). That sequence matters. Trying to model the entire gross distribution in one step gives the wrong answer because the stop-loss is triggered by aggregate outcomes, not individual large claims, and the per-occurrence layer is already removing the tail claims that would otherwise dominate the gross number.
A second, subtler wrinkle that Friedland calls out directly: net development factors can exceed gross when the fronting company retains the excess layer and the captive assumes the working layer (Friedland, p. 331). The intuition, if you have not seen it before, is that capping losses at the retention strips out the large-claim tail that makes gross development look heavy, so net is smaller in dollars but can develop faster in percentage terms. Board members who assume “net is always lighter than gross because reinsurance is risk transfer” will misread the report unless the actuary walks them through the specific structure.
What the board needs to see in the report is a clear bridge: gross losses of the underlying insureds, capped losses at the per-occurrence retention, and then the aggregate stop-loss adjustment on top. If any one of those three is missing, you cannot interpret the captive’s carried reserve. The numbers will look internally consistent because the spreadsheet balances, but the relationship between the captive’s book value and the actual risk being financed is opaque. The Johnson 2011 CLRS primer on captive basics for casualty actuaries is a useful reference for how group captive actuaries think about layered reinsurance.
The member allocation question
Most group captives allocate aggregate reserves back to individual members at each reserve study. The allocation drives dividend calculations, experience-rating adjustments for the next policy period, loss-pick reviews at renewal, and occasionally governance questions about whether a member should stay in the pool. Allocation is a practitioner convention rather than a method Friedland covers directly, so this section names practitioner convention explicitly rather than pinning a Friedland page number to material she does not discuss.
The allocation approaches you will see in practice fall into a few common buckets. Some captives allocate pro rata to premium, which is simple and auditable but unresponsive to member-level loss experience. Some allocate pro rata to reported losses, which rewards clean years and penalizes noisy ones, though it tends to double-count the noise from a single large claim. Some perform a mini loss-development exercise on each of the largest members and then allocate the residual to smaller members by formula, which is more actuarially defensible but more expensive. Hybrid approaches combine these elements with loss-corridor or experience-rating rules baked in.
The governance question every group captive board should be asking, and that most do not ask clearly enough, is this: does the sum of the member-level allocations reconcile to the actuary’s aggregate estimate? The two views should match, because they are describing the same liability. If they do not, the board should know why. A gap of one or two percent is usually an administrative rounding artifact or a difference in valuation date between the reserve study and the dividend-calculation spreadsheet. A larger gap almost always signals that one side or the other is using a different method, a different cohort, or a different set of assumptions, and the board needs to know which one is load-bearing for the financials.
A competent reserve study reconciles the two views in writing. If your study delivers the aggregate number and the member allocations in separate documents that never talk to each other, that is a governance gap worth closing before the next annual cycle.
What makes a group captive IBNR estimate go wrong
Friedland’s middle chapters systematically run each reserving method against the same portfolio under different kinds of operational change, and four of the failure modes are especially common in group captives. None of these is exotic; all are the predictable consequence of pooling.
TPA change. Group captives are a visible book of business, and third-party administrators compete aggressively to manage them. A TPA change can happen mid-cycle for cost reasons, service reasons, or because a member board lost confidence in the incumbent. When it happens, case reserving philosophy usually changes with it, and reported chain ladder becomes biased because recent reported dollars are not comparable to the pre-change history. The fix is the Berquist-Sherman adjustment to a consistent case adequacy level, applied to the reported triangle before projection. If your TPA changed during the experience period and the reserve report does not mention it, that is a gap.
Settlement speed shift. Paid chain ladder has the mirror problem. If claims are being closed faster under a new TPA or after a new settlement initiative, recent paid diagonals look more mature than the historical pattern and the method projects completion too quickly, understating ultimate. A slowdown does the opposite. The adjustment restates the paid triangle to a consistent disposal rate. Both kinds of shift can look like real improvements in the underlying loss picture when they are really just operational noise.
Change in member mix. A group captive that added a major member two years ago, or lost one last year, is reserving for a different pool than the historical triangle describes. Friedland’s general warning on a changing mix of business applies directly, and the fix is either to reweight historical data, apply expected claims to the current mix for recent years, or segment by cohort. Most group captive reserve reports handle member mix poorly, in part because membership changes feel like commercial information rather than actuarial information. They are both.
Large-loss leakage from the per-occurrence retention. If a large claim or two breached the captive’s per-occurrence retention and the layer is being paid out now, the capped triangle will understate ultimate unless the study explicitly adds a separate layer on top of the capped projection. This connects back to the previous section: the per-occurrence layer is doing real reserving work and has to be modeled, not assumed away. Friedland’s Berquist-Sherman advice on substituting policy year data for accident year data when limits or retentions change (Friedland, p. 283) is the standard adjustment when the retention itself has moved between policy years.
What the captive manager and member board should expect to receive
A reserve report that earns its fee delivers, at a minimum, six things in writing for a group captive or RRG.
An aggregate point estimate. The single number that goes on the captive’s balance sheet. ASOP 43 governs this and requires the actuary to produce an actuarial central estimate, defined as an expected value over the range of outcomes reasonably possible given the data and methods (Friedland, p. 10).
A reasonable range around the aggregate point. This is the window of estimates a second qualified actuary could defend on the same data, and it is the single best diagnostic of how confident the actuary actually is. The point estimate versus range explainer walks through the distinction for a non-actuarial reader.
A member-level allocation of the aggregate number, with a written reconciliation showing how the member pieces sum to the aggregate. See the previous section on why this matters for governance.
A Statement of Actuarial Opinion if the domicile requires one. Most U.S. domiciles require an SAO for licensed group captives, and every RRG is subject to annual actuarial opinion requirements as part of its statutory compliance, per NRRA’s comprehensive guide to the RRG structure.
A gross, ceded, and net bridge for every accident year, walking through the fronting and reinsurance structure in enough detail that a board member can follow it without a calculator.
A per-occurrence and aggregate stop-loss bridge showing how each layer is treated in the estimate. This is the piece that most reserve reports either hand-wave or bury in a footnote, and it is arguably the most important single line item for a member board member.
Cadence and cost
An annual Statement of Actuarial Opinion is the floor for licensed group captives and RRGs in most domiciles. NRRA’s guidance confirms that RRGs must submit annual actuarial opinions as part of their property and casualty filings, and most onshore group captive domiciles impose similar requirements through their captive statutes. The 2012 CLRS session on captive regulations and reserve certification walks through how the requirement plays out across domiciles in practice.
Some boards want more frequent information than once a year. Quarterly roll-forwards are common for larger captives, especially those with active dividend programs where the mid-year aggregate view drives a real decision. A roll-forward is typically a week of actuarial time, not a month, because the existing method selections are reapplied to a refreshed triangle rather than rebuilt from scratch. If your board wants quarterly visibility, it is almost always worth the marginal cost relative to the audit risk of discovering a material reserve miss at year end.
Engagement fees vary enough that any dollar figure here would mislead. Calibrate against the captive’s surplus and premium volume, not against a generic benchmark table.
Three concrete actions
If you have read this far and want to translate it into something your group captive or RRG can do this month, three moves will get you most of the value.
Ask your actuary for both the aggregate reserve and the member-level allocation, and verify the two reconcile. This is the governance question that most group captive boards never quite ask cleanly. A crisp reconciliation confirms the captive’s actuarial process and its dividend process are using the same data. A vague answer or a material gap tells you exactly what needs fixing.
Ask specifically how the per-occurrence layer and the aggregate stop-loss are being treated in the estimate. Friedland’s recommended sequence is to estimate ultimate claims limited to the per-occurrence retention first, then apply the stop-loss as a final adjustment (Friedland, p. 332). If your actuary cannot walk the board through that sequence in plain English, that is a board-level conversation worth having before the next renewal.
If membership has shifted materially in the last three years, ask how the historical data was adjusted for the change in mix. A member who joined two years ago or left last year distorts the historical triangle, and the fix (reweighting, expected claims on the current mix, or cohort segmentation) should be visible in the report. If it is not, the triangle is doing something the board cannot audit.
None of these steps requires you to become an actuary. They require questions your actuary should be happy to answer, and an independent second opinion is often the cleanest way to benchmark the answers you get back. The independent reviews note has more on why a second set of eyes matters for pooled structures specifically.