We now return to the IRS case involving Dr. Bernard T. Swift, Jr., where used a risk-pooled 831(b) captive insurance company, known as a microcaptive, to attempt to insure against the medical malpractice risks of 199 physicians in the employ of Dr. Swift’s Texas MedClinic. This was the subject of my previous article, Sixth IRS Victory Against Microcaptives In Swift (Feb. 10, 2024), and we return now to see how Dr. Swift fared in the U.S. Court of Appeals for the Fifth Circuit.
The opinion to be discussed is Swift v. CIR, Appeal No. 24-60270 (5th Cir., July 16, 2025), which you can and should read for yourself here. In a nutshell, Dr. Swift got walloped by the Fifth Circuit on pretty much every issue, with the Fifth Circuit affirming the U.S. Tax Court’s rulings ― including a 20% penalty. The only intrigue is why Dr. Swift decided to waste even more money on litigation fees to go to the Fifth Circuit in the first place since, to use a medical expression, this appeal was Dead On Arrival there.
Anyway, the Fifth Circuit mostly just confirmed and affirmed the findings of the U.S. Tax Court. There is no need to repeat all that analysis here and I will not bore you with it, but if you missed it the first time then you can gleam it all from my previous article or just read the Fifth Circuit’s opinion for yourself.
What we are going to talk about in this article is the concept of risk distribution as discussed by the Fifth Circuit and hopefully bring some clarity to that issue since all the courts seem to muddle it somewhat.
When it comes to captive insurance companies, risk distribution can be subdivided into two major parts. The first part is raw risk distribution which asks no more than whether the risk was distributed at all. The second part is risk distribution in the insurance sense which asks whether there was sufficient risk distribution that the captive could be considered an insurance company.
Raw Risk Distribution
The concept of raw risk distribution means that the potential for losses has been spread in a way that lessens the possibility of a total loss. This can be very minimal. Say that your loss is to come up tails in a coin toss. If you flip the coin once, there is no risk distribution since it is only heads or tails ― you will either be a winner or loser. Your risk is 50/50. That is just a gamble or wager.
Flip the coin twice and now there is an additional possibility that one flip will result in a head and one flip will result in a tail. So now your risk of loss looks like this:
Head : Head
Head : Tail
Tail : Head
Tail : Tail
This means that your risk of the worse loss (tail : tail) is only 1 in 4 our 25%. Thus, by adding an extra coin flip, you have reduced your risk by half (50% to 25%) and thus distributed your risk over two flips instead of one. This is raw risk distribution in its most simple form and it is no more complicated than that.
Risk Distribution In The Insurance Sense
The more times that you flip the coin, the more that your risk of total loss decreases. So, let’s say that you flip the coin 20 times. While it is possible that the coin could come up tails 20 times in a row, the odds of that happening are so tiny that you would doubtless want to inspect the coin more closely afterwards.
But let’s say that you flipped the same coin a 100 times. By this point, the odds of the coin coming up tails a 100 times is almost (but not quite) non-existent. To the contrary, you would start to see something like an even balancing of the heads and tails, probably not an exact 50/50 distribution but within the same ballpark. This is where we start to see something known as the Law Of Large Numbers, which is not a legal concept for a statistical one.
The Law Of Large Numbers effectively posits that if you flip the coin a large number of times then the result becomes statistically predictable and subject to only slight statistical deviations which themselves can be accurately predicted. In other words, if you flip the coin a million times then you can expect something very close to a 50/50 distribution of heads and tails. The outcome thus becomes reasonably predictable.
The concept of risk distribution in the insurance sense starts with the predictability of the Law Of Large Numbers. As the Fifth Circuit related in its Swift opinion:
“The law of large numbers tells us that when there are a sufficiently large number of independent risks each having an annual probability of X%, there is an extraordinarily small likelihood that the percentage of insureds that suffer a loss during a year will deviate significantly from X%. Put differently, if a coin is tossed a million times, it is highly unlikely that the percentage of heads will differ appreciably from 50%. In this way, insuring a large number of independent risks protects the insurer against financial calamity, because the insurer can accurately predict losses for the group as a whole, and set premiums accordingly. Therefore, risk distribution depends on the existence of a sufficient number of independent risks.” [Citations and inner quotation marks omitted.]
Why is this predictability so important? It is not just the threat of financial calamity ― our coin coming up tails 10 times in a row ― that must be considered. Instead, predictability underpins the entire business of insurance.
The sine qua non of any business is to generate profits. To run an insurance business profitably means that the insurance policies issued by the company need to be net profitable in the aggregate. To make this happen, at least three things have to be taken into account.
First, like any other business, there will be expenses in running the business. An insurance company, even a captive, incurs costs for doing various things, such as underwriting policies, administering claims, managing the company, and the like. Next, there are other expenses of the insurance company such as licensing fees.
Second, there are the insurance risks. Every insurance policy is a gamble: The policyholder buys the policy because she thinks some event might happen, and the insurance company sells the policy thinking that same event might not happen. But nobody has a working crystal ball or a special OUIJA board that can accurately say in advance that the event will happen or not. But here the insurance company has an edge, since it can distribute its risk among policyholders such that the odds of the coin flip coming up tails for all the policyholders becomes almost zero.
Third, the insurance policies sold by the insurance company need to be profitable in the aggregate. Oh, sure, the insurance company can expect to get walloped on some policies, but on others it will take in premiums and never have to pay out our flipped penny. Thus, if the insurance company can sell enough policies, then the Law Of Large Numbers kicks in and the insurance company can then predict that if it charges $X per policy that it will realize $Y in profits.
So there you have it: An insurance policy is priced largely according to three components: Expenses, losses, and profits. Thus, if the insurance company’s underwriter anticipates that the expenses of a given policy will be $200, the losses spread over all policies will be $2,000, and the company needs $50 off each policy to be profitable, then the underwriter will price the policy at $2,250. Of course, the underwriter will try to get more for each policy if the market supports that, but he would be stupid to write that policy for less than $2,250.
Note that in this calculation, only two elements are fixed at what they actually need to be: Expenses and profits. The wildcard is the insurance risk since it cannot be predicted in advance. The problem is found in the deviations, which are sometimes called outliers. Going back to flipping the coin a million times, in the end one can expect something very close to a 50/50 distribution, but somewhere in the middle a group of flips may come out 80 heads and 20 tails. If an insurance policy is underwritten for only those particular 100 flips, the insurance company could suffer substantial losses on that policy.
This is what happens, for instance, with wind and flood policies sold in Florida. The insurance company’s actuary may look at a bunch of data, not the least of which being the frequency of hurricanes, and make the prediction that only one major hurricane would hit the sunshine state every 10 years. The insurance company’s underwriter takes this data and prices policies on the expectation that in any given year, the odds of a majority hurricane arriving is only 1 in 10. The underwriter would be additionally comforted that the policies are distributed all over the state and not just in one area. But let’s say it is a really bad year for hurricanes and not just one, but two hurricanes show up that year, with one hitting Miami and the other hitting Tampa. Now the insurance company takes a major loss even though the actuary’s predictions were quite reasonable based on the data.
As an aside, it is for the threat of such outliers that the insurance companies that sell insurance policies directly to purchasers (sometimes known as retail insurance companies) will often hedge or lay off their risks by purchasing their own insurance through what are known as reinsurance contracts. Reinsurance shifts some or all of an insurance company’s loss exposures to another insurance company who is willing to take that risk on … for their own expectation of profits.
The point being that an insurance company must have sufficient resources, or offload some of its risks through reinsurance contracts, that it can withstand reasonable loss outliers and still survive. At the same time, the insurance company must be able to distribute its risks over such a large number of policyholders that its risks of a calamitous loss are minimized. This latter part is what “risk distribution in the insurance sense” is all about.
To say that the U.S. Tax Courts and the reviewing U.S. Courts of Appeals have struggled with this issue, and continue to struggle with this issue, would be somewhat of an understatement.
Here must digress to a famous (or infamous) decided by the U.S. Supreme Court involving pornography, being Redrup v. New York, 386 U.S. 767 (1967). The Redrup opinion came in the midst of a number of U.S. Supreme Court opinions around this time when pornography was largely still illegal and the justices were struggling to legally define it. The holding of Redrup was basically that each justice would review the materials being sold by a particular defendant and then each would decide according to their own understanding whether the materials were indeed pornographic. This process became known as “Redrupping”― basically, not stating a cogent standard that anybody could abide by, but instead looking at the evidence and making a purely subjective determination. That is not what courts are supposed to do, and in 1973 the Redrup decision was thrown out by the U.S. Supreme Court precisely because it was so subjective.
The process of Redrupping cases ― not stating a standard but looking at each case subjectively ― is what has going on with the U.S. Tax Court and the U.S. Courts of Appeals that have been reviewing issues of risk distribution in the insurance sense. We can discern this from the following paragraph in the Swift opinion:
“Finally, the number of independent risks ensured by the Captives is ‘at least a couple orders of magnitude smaller than the captives in cases where [the tax court] found sufficient distribution of risk.’ Swift, 2024 WL 378671, at *18 (quoting Caylor Land, 2021 WL 915613, at *12); see Rent-A-Center, 142 T.C. at 24 (captive provided workers’ compensation, automobile and general liability insurance for 14,300 to 19,740 employees, 7,143 to 8,027 vehicles, and 2,623 to 3,081 stores); Securitas Holdings, 2014 WL 5470747, at *9–10 (captive provided workers’ compensation, automobile and general liability insurance to 25 to 45 entities in over 20 countries, covering over 200,000 employees and 2,250 vehicles).”
This passage is little more than the Fifth Circuit saying, “the numbers of those other cases are big enough to provide risk distribution, and this is not one of those cases.” It provides no guidance to as to when the number of independent risks (also called points of risk or sometimes risk points) is large enough to qualify as risk distribution nor small enough that they do not. All we really come away with is that this particular U.S. Tax Court judge and particular panel of the Fifth Circuit subjectively thought that it wasn’t enough here. This is Redrupping in a nutshell and the more quickly this “reasoning” is abandoned by everybody involved the better. Anytime the courts end up in the logical cul-de-sac where they say that “4,000 is enough but 3,999 isn’t” then they have made a wrong turn and need to reverse course.
The better analytical construct is to do away with this counting of risk points altogether and instead look to whether the risk distribution was sufficient that profits and losses from all policies written in the aggregate could be reasonably predicted. In other words, there is sufficient risk distribution that the Law Of Large Numbers comes into play such that the outcome would be within a reasonable standard deviation for insurance companies generally. This would not require the setting of a specific number ― which would be impossible in the insurance context anyway because risks and coverages can vary so greatly ― but it would allow captive owners, the IRS, and reviewing courts a decent framework to determine whether risk distribution is present.
Not that a better framework would have changed the result in this case since there were so many other things fundamentally wrong with the captive arrangement. But it might be useful in future cases involving points of risk somewhere between these microcaptive tax shelter cases and the large corporate captives that have many thousands of diverse risks being insured. It would at least get rid of the wholly-subjective Redrupping approach currently being used which ends up being wholly subjective and substantially arbitrary, which is anathema to the legal system.