Risk distribution, also known as risk sharing, is a fundamental feature of insurance. I think that the best definition of risk distribution is this: The (actuarially credible) premiums of the many pay the (expected) losses of the few. This is the essence of insurance.
Single-parent captives that cannot spread their parents' risks over a pool of disparate corporate entities are not captives; they're financial vehicles into which the parents have transferred funds for the purpose of paying their insurable losses. As such, the parents of such financial vehicles cannot deduct their payments (they're not premiums because the vehicle is not an insurer) from their U.S. federal income taxes.
Homogeneity of Loss Exposures
This is as far as most people go in their interpretation of Revenue Ruling 2005–40; spread the risk among at least 12 independent entities and risk distribution ensues. However, there is another aspect of this revenue ruling that many people miss or ignore—homogeneity of loss exposures. Each of the four situations in the revenue ruling state that the risks in question are homogeneous; they are the same line of insurance, and the same line of business. For example, a fleet of 100 tractor-trailer trucks that traverse the country each week are each subject to the same type, frequency, and severity of automobile liability losses.
Many captive practitioners disagree with the notion that in order to attain risk distribution the risks must be homogeneous. Their argument is based on portfolio theory. They reason that heterogeneous groups, by virtue of their uncorrelated loss characteristics, produce risk distribution. It is true that a portfolio of loss exposures is, to some degree, mathematically predictable due to the portfolio's internal hedging—when a loss occurs in one line or one company, there are two or three others that do not produce a loss.
While this is true, the portfolio effect is not the basis for calculating expected losses. The primary requirement for the calculation of expected losses is a large pool of homogeneous risks; this is what transforms a loss funding vehicle into a bona fide insurance company. Therefore, only a large number of homogeneous risks create risk distribution. There is no other way to interpret Revenue Ruling 2005–40.
For example, a group of 10 middle-market bakeries exhibit similar exposures to workers compensation losses, and if the collective historical loss data are actuarially credible, i.e., are large enough, expected losses, premiums, rates, and surplus requirements may be reasonably determined in order to form a captive. In these types of captives, a safe premium-to-surplus ratio may be between 5 to 1 and 3 to 1.
Now, add to this group of bakeries the workers compensation risks of 2 valve manufacturers, 1 cement contractor, 3 utilities contractors, and 2 banks. We now have a heterogeneous group with 18 members, and so the prevailing argument goes, the loss outcomes for the portfolio should be more predictable than that of the portfolio with only 10 homogeneous risks, because of the noncorrelation among the various workers compensation loss profiles. This is true, noncorrelation reduces volatility, but reducing the portfolio's volatility does not create risk distribution—only the ability to determine expected losses, rates, premiums, and surplus, in other words, form an insurance company, creates risk distribution.
In this example, the eight additional risks do nothing to enhance the loss predictability of the bakeries, and unless the others are very large companies, they cannot generate loss predictability on their own. (If they were very large, they wouldn't be joining a group of middle-market bakeries.) Actually, these eight additional risks decrease the actuary's ability to calculate expected workers compensation losses for the group. In this case, the funding requirements would have to be very high, thus rendering the captive uncompetitive and unattractive, even to the bakeries. This is because the portfolio effect does not create risk distribution.
Not only are the group's losses not reasonably predictable in this scenario, the surplus requirements cannot be as easily determined as in the example with just the 10 bakeries. In fact, in this scenario, we would be negligent if we used the above solvency benchmarks, as the workers compensation risk profile of the contractors and manufacturers are far different than that of the bakeries. Some might say that the banks' relatively low loss probability would offset the others' losses. That might be true, but we still cannot accurately calculate expected losses for the group, which is the primary objective for an insurer, so we still do not have risk distribution as prescribed in Revenue Rule 2005–40.
The point to remember is that expected losses can only be reasonably determined when the risk pool has enough homogeneity, and the goal of risk distribution is the ability to predict future losses so rates, premiums, and surplus requirements can be reasonably determined. This is done solely on a per-line of business basis. Large commercial insurers need not worry about this, of course, as they have actuarially credible pools of contractors, manufacturers, and bakeries for each line of coverage they offer. Captives, on the other hand, cannot rely on portfolio theory to substitute for risk distribution.
An excellent example of why homogeneity is required for risk distribution (and therefore future loss predictability) is the National Council on Compensation, Inc. (NCCI). The NCCI gathers loss data for a large number of industries, and establishes loss rates for each. Workers compensation rates are a combination of loss costs and expenses. (Workers compensation insurers generally add the expense component, but the NCCI will do so for several states.)
The NCCI loss costs are based on the enormous homogeneous pools of data generated by every industry. If the portfolio effect produced risk distribution, the NCCI could simply lump all industry loss data into a single pool, creating a huge number of uncorrelated risks, and every employer in the 35 NCCI states would pay the exact same workers compensation loss cost rate.
The portfolio effect is useful in helping to determine proper surplus requirements for a captive with several lines of actuarially credible pools of homogeneous risks. For example, Captive A covers its parent's subsidiaries' workers compensation, general liability, and long-term disability risks. The parent is a large multinational company, with a large number of independent subsidiaries, so each line of insurance has enough homogeneous exposure units spread among a large group of corporate entities to calculate expected losses (risk distribution).
We can prove mathematically that a portfolio (captive) with three lines require less surplus than would each line of insurance individually. If none of the three lines had enough exposure units to create risk distribution, the portfolio effect might still be somewhat valuable in determining the surplus requirements, but it would not create risk distribution.
Finally, I recognize that the portfolio effect can be a powerful ally in managing a multiline captive's loss volatility. Uncorrelated risks, as with uncorrelated investments, in a portfolio produce better, more predictable outcomes than do the simple sums of the risks or investments. But as every actuary knows, the loss predictability that results from the portfolio effect, while a very good thing, is not a substitute for the predictability calculated from a large amount of homogeneous loss data. And it's the ability to predict expected losses that creates risk distribution.