Captive funding at a confidence level is the practice of capitalizing the captive’s reserve account at the percentile of the aggregate loss distribution chosen by the buyer. At the 90th percentile, the funded amount covers actual losses in nine years out of ten; in the tenth year, the captive’s losses exceed the funded amount and additional capital is needed. The phrase is captive-industry shorthand for a specific quantitative exercise that determines how much capital the captive holds against its expected liabilities, and the chosen percentile is one of the most consequential single decisions in the captive’s annual financial plan.
Three things about the practice trip up first-time captive owners. The percentile is not a confidence interval in the statistical sense. The variance assumption that drives the aggregate distribution can move the funded amount more than the percentile selection does. And the funded amount is not the same thing as the regulator’s required capital, and the gap between the two is itself a decision the captive must make.
This article walks through what confidence-level funding actually is, how the actuary builds the aggregate distribution that supports it, why the variance input matters more than most readers realize, what the buyer’s decision actually is, and what to require from the actuary’s report on the funded amount. The starting points are the actuarial central estimate for the difference between a single estimate and a distribution, the captive feasibility study for the formation-stage version of this analysis, and the single-parent captive guide for the standard reserving framework the funding analysis sits inside.
What confidence-level funding actually is
The captive holds capital and reserves against its expected loss liabilities. The expected loss is the central estimate the actuary produces under ASOP 43, defined as an expected value over the range of outcomes reasonably possible (Friedland, p. 10). The central estimate is one number. Actual losses in any year will be higher or lower; the distribution of possible outcomes spreads above and below the central estimate.
The funded amount is the dollar number the captive holds to absorb realized losses up to a chosen percentile of the distribution. At the 80th percentile, the funded amount covers losses in eight years out of ten. At the 90th percentile, nine years out of ten. At the 99th percentile, ninety-nine out of one hundred. Higher percentiles produce larger funded amounts and tie up more capital, but reduce the probability that the captive will need a mid-year capital call to make claim payments.
The confidence level is not a statistical confidence interval. A 90 percent confidence interval is a range constructed so that, with 90 percent probability, the true mean lies within the range. The 90th percentile of an aggregate loss distribution is a single dollar amount such that, with 90 percent probability, actual losses fall at or below it. The two concepts use overlapping language and are routinely confused; the distinction matters because the percentile-based funded amount has a specific operational meaning that the confidence-interval framing does not.
Friedland defines the central estimate but does not prescribe a confidence level for capital. The choice is the captive’s, made by the board with the actuary’s modeling as input. The choice has direct financial consequences: capital held inside the captive is capital not deployed elsewhere. Setting the percentile too high ties up parent organization capital for no economic return. Setting it too low raises the probability of a mid-year capital call and the disruption that comes with it.
The aggregate distribution
The funded amount is the percentile of a distribution. The distribution does not arrive ready-made; the actuary constructs it from three building blocks.
Severity distribution. The actuary fits a parametric distribution to the severity of individual losses on the captive’s book. Lognormal is the default for most liability lines. Gamma fits some workers compensation severity patterns. Pareto and the related heavy-tail families fit lines with a meaningful right tail, including general liability, hospital professional liability, and any line with a meaningful risk of nuclear verdict severity. The choice of family is itself a judgment, and the actuary should document why the chosen family fits the captive’s specific loss data.
Frequency distribution. The actuary fits a parametric distribution to the count of claims per period. Poisson is the textbook default and fits most lines with stable annual frequency. Negative binomial accommodates overdispersion (variance larger than the mean), which is common in lines where some years produce clusters of claims and others produce few. The choice between Poisson and negative binomial can affect the right tail of the aggregate distribution meaningfully when frequency variance is high.
Combination. The severity and frequency distributions are combined into the aggregate loss distribution through one of three methods. Analytical convolution (closed-form combination of the two distributions) is feasible only for specific severity-frequency pairings and is rarely used in practice for captive work. Monte Carlo simulation (drawing N severity samples per period across many simulated periods) is the dominant practical method, run with 10,000 to 100,000 iterations to produce a smooth empirical distribution. Fast Fourier transform methods are an analytical alternative that converges to the same answer as a simulation with enough iterations.
The aggregate distribution’s central tendency is approximately the central estimate; the actuary should reconcile the simulation mean to the central estimate within tolerance. The percentile of interest is read off the cumulative distribution function of the simulation output. The 90th percentile is the simulation output’s 90th-percentile dollar figure; the 80th, the 80th; and so on. The funded amount the buyer holds is read off the curve at the buyer’s chosen percentile.
The choice of severity family, frequency family, and the parameters fit to each are all actuarial judgments. The same loss data fit by different family choices can produce different percentile estimates, sometimes by a meaningful amount on the right tail. The actuary should document each choice explicitly. The buyer should be able to see, at the report level, which severity distribution was selected and why.
Friedland’s frequency-severity framework (Friedland, p. 195) is the foundation. The funding analysis takes the same frequency-severity modeling further into the right tail, where the captive’s capital adequacy is determined.
Why the variance assumption drives the answer
The percentile selection is the input most discussed at the board level. The variance assumption is the input that most often dominates the answer.
Two captives with the same expected loss can produce funded amounts at the 90th percentile that differ by a factor of two if their underlying severity variance assumptions differ. A workers compensation captive whose severity is well-described by a lognormal distribution with a moderate coefficient of variation produces one funded amount. A general liability captive whose severity follows a heavier-tail Pareto-style distribution, with the same expected severity but a much fatter right tail, produces a funded amount that can be twice as large at the same percentile.
The intuition is direct. The 90th percentile of a thin-tailed distribution is close to the mean. The 90th percentile of a fat-tailed distribution is far above the mean. The two distributions may have the same expected value, but their distance from the mean to the 90th percentile is different, and the captive’s funded amount is read off that distance.
The variance input has three components the actuary must choose. The variance of the severity distribution (controlled by the severity family choice and the dispersion parameter). The variance of the frequency distribution (controlled by the Poisson-versus-negative-binomial choice and the overdispersion parameter). The correlation between severity and frequency, which is often assumed zero but in some lines is meaningfully positive (years with high frequency tend also to have high average severity).
Each component requires data to fit, and the captive’s own data is often thin. The actuary’s reliance on industry benchmark variance estimates is therefore a structural feature of the analysis, not a deficiency. Friedland’s benchmark caution (Friedland, p. 88) applies to the variance estimate as forcefully as it applies to the expected estimate, and the buyer should ask the same questions about variance benchmark comparability that the buyer asks about expected-loss benchmark comparability.
The practical implication for the buyer is that a sensitivity analysis on the variance assumption matters as much as a sensitivity analysis on the percentile selection. A report that varies the chosen percentile from 80 to 90 to 95, holding the variance fixed, tells the buyer one dimension of the uncertainty. A report that also varies the variance assumption, with the percentile fixed, tells the buyer the dimension that often matters more.
The percentile selection itself
The percentile is the buyer’s decision. Three considerations frame the choice.
The captive’s purpose. A captive that is a primary insurer for the parent organization’s most consequential exposures needs higher percentile funding than a captive that is a small piece of a larger risk-financing program. The 90th percentile is common for primary captives; the 75th to 80th is common for cells or for smaller captives operating alongside commercial insurance.
The capital cost. Capital held in the captive is capital not deployed in the parent’s operations. The parent’s internal cost of capital sets a ceiling on how much it is willing to hold against captive losses. A parent organization with a 10 percent internal cost of capital effectively pays 10 percent annually on every dollar held in the captive’s surplus that could have been deployed elsewhere. The percentile selection has a direct cost; the higher the percentile, the higher the foregone return.
The probability of mid-year capital call. Setting the percentile too low raises the probability that the captive will need a mid-year capital injection from the parent. The disruption from a mid-year call is operational (the parent must move capital and the captive must accept it), regulatory (the domicile may scrutinize the captive’s adequacy more closely after a call), and reputational (the captive’s posture in the eyes of its reinsurers and rating agencies softens). For a parent organization that prioritizes predictability, the higher percentile is worth the foregone return; for a parent organization that prioritizes capital efficiency, the lower percentile is worth the call risk.
A typical decision falls between the 75th and the 95th percentile, with 80 to 90 the most common range. Selections outside that band are common only in specific structural situations: cells operating at minimum statutory capital may fund at lower percentiles (closer to 65 or 70); risk retention groups subject to additional regulatory scrutiny may fund at higher percentiles (95 and above). The board’s role is to make the selection on the basis of the actuary’s distribution, the parent’s cost of capital, and the operational consequence of a call.
Funded amount versus regulatory required capital
The funded amount is the captive’s choice. The regulatory required capital is the domicile’s floor. The two are not the same number, and the gap between them is a decision.
Most U.S. captive domiciles set minimum capital based on a combination of statutory floor (a fixed dollar amount tied to the captive’s structure) and risk-based capital (a formula tied to the captive’s lines of business, reinsurance arrangements, and asset mix). The formula approach produces a regulatory number that is unrelated to the captive’s specific loss distribution, the buyer’s chosen percentile, or the variance assumption. The regulatory number is the floor below which the regulator may take action; it is not a recommendation about the appropriate funded amount.
For most captives at most maturities, the funded amount at the buyer’s chosen percentile exceeds the regulatory floor. The captive’s surplus position should be set with the actuary’s distribution in view, not the regulator’s formula. A captive whose surplus only meets the regulatory minimum is operating at a lower probability of solvency than most buyers think; a captive whose surplus meaningfully exceeds the regulatory minimum may be operating at a higher percentile than the buyer intended.
The captive’s annual financial plan should reconcile the three numbers: the central estimate of the reserve, the funded amount at the buyer’s chosen percentile, and the regulatory required capital. If the three are not all visible in the captive’s planning documents, the captive is operating without a clear view of its own capital adequacy. The audit committee governance frame puts this reconciliation on the board’s reading list at the right cadence. The captive domicile survey covers how the regulatory floor differs across the major domiciles.
How confidence-level funding differs from the central estimate and the range
Three numbers appear in any well-documented captive reserve analysis. They are related but distinct, and the distinctions matter.
The central estimate. The actuary’s single best answer for the expected ultimate loss. Defined under ASOP 43 as the expected value over the range of outcomes reasonably possible (Friedland, p. 10). This is the number the captive books on its balance sheet as the reserve.
The reasonable range. The window of estimates a different qualified actuary could defend as reasonable on the same data. This is not a statistical confidence interval; it is the range of methodology choices and judgment selections a competent actuary could make. The framing is described in point estimate versus range for a non-actuarial reader.
The funded amount at a confidence level. The percentile of the aggregate loss distribution the captive chooses to capitalize against. This number is typically the largest of the three for most captives, because the percentile is set above the central estimate and the right tail of the distribution can be meaningful.
The three numbers answer different questions. The central estimate asks: what is the expected ultimate loss. The range asks: how wide is the actuarial professional disagreement on the expected ultimate. The funded amount asks: how much capital does the captive hold to absorb a year in which actual losses are higher than expected.
A report that presents only one of the three is incomplete. The reserve booking decision keys off the central estimate. The professional reasonableness review keys off the range. The capital adequacy decision keys off the funded amount. Each decision needs its number; the report should produce all three explicitly and reconcile them.
Sensitivity testing on the funded amount
A funded amount with no sensitivity analysis is not useful. The buyer cannot evaluate how the number was produced, cannot judge how the number would move under plausible alternative assumptions, and cannot calibrate how much margin the funded amount carries.
A reasonable sensitivity analysis varies four inputs.
The percentile selection. Show the funded amount at the 75th, 80th, 85th, 90th, and 95th percentiles. The shape of the curve communicates how sharply the funded amount climbs as the percentile rises.
The severity variance. Show the funded amount at the central severity variance, at a 25 percent higher variance, and at a 25 percent lower variance, holding the percentile fixed. The width of the band communicates how exposed the funded amount is to the variance assumption.
The frequency variance. Show the funded amount under Poisson, under moderate-overdispersion negative binomial, and under high-overdispersion negative binomial. The differences communicate the impact of the frequency-variance choice.
The tail factor. For lines where the tail factor is a meaningful input, show the funded amount at the central tail, at a 10 percent higher tail, and at a 10 percent lower tail.
The four-by-four-by-four grid is impractical to display in full. The actuary’s report should pick the two or three inputs that move the funded amount most and show the joint sensitivity to those. A tornado chart or equivalent display is the standard. The buyer should know, by the end of the sensitivity section, which two or three assumptions are doing most of the work on the funded amount.
Year-over-year monitoring
The funded amount is not a one-time decision. The captive’s loss distribution evolves; the percentile selection should be revisited as the captive matures and as the loss experience accumulates.
Three drivers move the funded amount year over year, holding the percentile fixed.
Maturing data. A captive in its first three years relies heavily on industry benchmark variance estimates. As the captive accumulates its own loss data, the variance estimate transitions from benchmark to own data. The transition usually narrows the variance (own data on stable operations typically shows less dispersion than industry aggregate data), which reduces the funded amount at the chosen percentile.
Operational changes. A material change in the captive’s exposure (a new line, a new geographic territory, a new acquisition by the parent) changes the loss distribution. The variance assumption needs to be refit, and the funded amount needs to be revisited at the same percentile.
Tail factor drift. For long-tail lines, the tail factor selection itself drifts as new diagonals of data emerge. A change in the tail factor flows through the expected loss and into the variance of the distribution, and the funded amount moves accordingly. The frame in tail factor selection describes how the tail input moves and why.
The captive’s annual reserve study should produce a year-over-year comparison of the funded amount at the chosen percentile, with explanations for any material change that is not driven by paid claims or new business. The frame in interim monitoring applies; mid-year drift in the loss distribution can affect the funded amount before the next annual study formalizes the change.
How funding interacts with LPTs and ADCs
The funded amount the captive holds is a current statement of capital adequacy. A material transaction during the year can change the underlying loss distribution and therefore the funded amount.
A loss portfolio transfer removes a block of reserves from the captive and replaces it with a reinsurance recoverable. The aggregate loss distribution shrinks to reflect the removed block. The funded amount falls, releasing capital that had been held against the transferred reserves. The release is one of the principal economic motivations for the transaction.
An adverse development cover does not remove the reserves but caps the right tail above the attachment. The aggregate distribution is truncated at the limit of the cover. The funded amount at a high percentile can fall meaningfully, because the right tail above the attachment is now bounded by the cover rather than unbounded. The release is smaller than an LPT release but real, and the cost of the cover may be lower than the cost of an LPT for equivalent capital release.
The funding analysis should be rerun after any material transaction. A captive that completes an LPT or ADC and then continues to hold capital at the pre-transaction funded amount is operating at a higher percentile than the chosen one, which may be the right answer for some boards but should be an explicit decision rather than a passive result.
What to require from the actuary’s report
The buyer’s checklist on confidence-level funding analysis is concrete. Each item below is reasonable to expect.
The severity distribution family selected and the parameters fit, with a goodness-of-fit diagnostic against the captive’s own data. The frequency distribution family selected (Poisson or negative binomial), with the overdispersion parameter if applicable.
The aggregate distribution presented in summary statistics (mean, standard deviation, coefficient of variation, skewness) and at named percentiles (50th, 75th, 80th, 85th, 90th, 95th, 99th).
The funded amount at the buyer’s chosen percentile, with the central estimate and the reasonable range presented alongside for comparison.
A sensitivity table varying the percentile, the severity variance, the frequency variance, and the tail factor independently, with the joint sensitivity to the two or three most influential inputs presented as a separate display.
A reconciliation against the regulatory required capital, with the gap explained. If the funded amount exceeds the regulatory minimum, the gap is the buyer’s margin and the report should say so. If the funded amount falls below the regulatory minimum, the buyer has a problem the captive’s planning must address.
A year-over-year comparison of the funded amount at the chosen percentile, with explanations for material movement that is not driven by paid claims or new business.
The variance benchmark sources used (industry, prior parent experience, current captive experience) with comparability disclosure of the kind Friedland describes for expected-loss benchmarks (Friedland, p. 88).
A signed actuarial opinion on the reasonableness of the funded amount at the chosen percentile, under the standard that applies to the captive’s domicile.
A report that omits any of these items has a documentation gap. The diligence frames in evaluating the actuary’s report and reading the actuarial proposal apply to the funding analysis with the same force they apply to the reserve analysis.
The board’s role on the funded amount
The funded amount is the single most consequential capital decision the captive board makes annually. The decision sits at the intersection of the actuary’s distribution, the parent’s cost of capital, and the buyer’s tolerance for a mid-year call. The board should ask three questions every year.
What is the funded amount at the chosen percentile, and how does it compare to the central estimate of the reserve and to the regulatory required capital?
What is the variance assumption driving the funded amount, and where did that assumption come from? If the variance is benchmark-derived, what does the captive’s own data say about whether the benchmark is comparable?
What is the year-over-year change in the funded amount, and what is driving it? If the change is material and is not driven by paid claims or new business, the change is itself a finding that needs explanation.
The funded amount is the captive’s expression of how much capital it holds against uncertainty. The percentile is the buyer’s tolerance for that uncertainty made operational. The variance assumption is the actuary’s expression of how wide the distribution is. The three together determine the captive’s capital adequacy; the captive operates as well as those three decisions allow it to.