Unum Life Insurance Company v. United States

897 F.2d 599, 12 Employee Benefits Cas. (BNA) 1084, 65 A.F.T.R.2d (RIA) 768, 1990 U.S. App. LEXIS 3204, 1990 WL 19707

• Court: Court of Appeals for the First Circuit
• Decided: March 6, 1990
• Docket: 89-1454
• Status: Published

Opinion

BREYER, Circuit Judge.

The question on appeal is whether certain amounts that UNUM Life Insurance Company (“UNUM”) maintains to safeguard the payment of its obligations under “deposit administration contracts” (containing “annuity options” with “permanent rate guarantees”) qualify as “pension plan reserves” within the meaning of § 801(b)(1) of the Internal Revenue Code of 1959, Pub.L. No. 86-89 § 2(a), 73 Stat. 112, repealed and replaced by Pub.L. No. 98-369, Div. A, Tit. II § 211(a), 98 Stat. 721 (current version at §§ 801-818 of the Internal Revenue Code of 1986, 26 U.S.C. §§ 801-818), which old version we shall still refer to as the “Code.” If not, UNUM can deduct from its taxable income (for 1977 and 1978) the actual amount of “interest” that it paid on money deposited under those contracts; if so, UNUM can deduct only the amount deposited multiplied by the interest rate that UNUM itself earned, on average, on its own investments — an amount that turns out to be somewhat smaller. The district court, 709 F.Supp. 13, held, on the basis of the factual record, that the relevant amounts did not meet the statutory definition of “pension plan reserves.” See IRC § 801(b)(1). (This provision, and others cited throughout, were repealed in 1984). It therefore allowed UNUM the larger deduction. After re viewing the record and the relevant law, we agree with the district court.

I

Background

Before turning to the specific legal question presented, we shall briefly describe, sometimes in a highly oversimplified way, some of the relevant legal and factual background.

1. The taxation of a life insurance company’s investment income. The tax law recognizes that life insurance companies may engage in several different kinds of business, selling or administering, for example, ordinary life insurance, various sorts of pension plans, and other kinds of retirement programs. In most instances a buyer will pay the company money (often, but not always, called a “premium”) and the company will invest this money, earning a return. The company uses the total amount of money it obtains from both premiums and investment income to pay (1) benefits and (2) administrative expenses. The remainder is profit.

Since the company may pay out, in the form of benefits, much of the return that it earns on its investments, one cannot reasonably treat that return as if it were the company’s own income. Yet, some of that money may well end up in pockets of shareholders, particularly if the company (perhaps following state regulators’ properly conservative accounting practices) has put aside (as a reserve) considerably more money to pay future benefits than it actually needs. Consequently, the Internal Revenue Code uses a fairly complex formula to divide the total return that the life insurance company earns on its investments each year into two parts, one part that it considers as earmarked for the payment of future benefits (a beneficiaries’ share) and another part that it considers likely to end up as company profit (a company share). The Code permits the company to deduct the beneficiaries’ share from its net investment income before determining the investment income upon which it must pay tax. See IRC § 804(a)(1). The way the Code calculates the beneficiaries’ share of investment income depends upon the type of business, and consequently the kind of beneficiary, involved. In the case of ordinary life insurance, for example, the Code multiplies the “adjusted life insurance reserves” (the life insurance reserves the company would hold if it calculated them using realistic, rather than very conservative, interest rate assumptions) by the “adjusted reserves rate” (the company’s average rate of return on all its assets over one or more years) to obtain the amount of additional money the company had to add to its reserves during the past year to maintain an amount sufficient to pay future benefits. See IRC § 805(a)(1), (b, c). In the case of “pension plan reserves” (which are special kinds of life insurance reserves), see IRC § 805(d), the Code multiplies the average amount of such reserves held during the year by the average amount that the company earned on all its investments during the year, without “adjusting” either amount. See IRC § 805(a)(2), (b)(2), (d). In the case of amounts held under other types of contracts (highly relevant here), the Code uses the actual amount of interest that the company credited to the reserve. See IRC § 805(a)(3), (e). The company then adds up all these amounts and arrives (roughly speaking) at the beneficiaries’ share of the company’s investment income.

To understand the basic idea, imagine, for example, that a life insurance company has assets of $1 billion and earns $150 million (or 15%) in a given year by investing those assets. If its “adjusted life insurance reserves” amounted to $200 million and the “adjusted reserves rate” was 12%, it would multiply 127° times $200 million, yielding $24 million. If it maintained another $300 million of “pension plan reserves,” it would multiply its 157° average earnings rate times $300 million, yielding $45 million. If it had an additional $400 million earmarked to pay other benefits (related, say, to other kinds of retirement plans), it would take the actual interest it credited to those “accounts,” let us imagine $31 million. It would then add these three figures together, yielding a total of $100 million. This figure of $100 million repre sents (roughly speaking) the beneficiaries’ share of the company’s total investment earnings of $150 million. The company may (roughly speaking) deduct this $100 million from its net investment income in order to obtain its investment income for tax purposes.

In fact, the Code’s actual method for calculating the tax-free portion of investment income is slightly more complicated than we have just indicated. The calculations we have described do not actually result in a deductible beneficiaries’ share; instead, they result in an amount called the “policy and other contract liability requirements.” The complete formula appears in section 805(a) of the Code, which says that the “policy and other contract liability requirements” are the sum of—

(1) the adjusted life insurance reserves, multiplied by the adjusted reserves rate,

(2) the mean of the pension plan reserves at the beginning and end of the taxable year, multiplied by the current earnings rate, and

(3) the interest paid.

I.R.C. § 805(a). Once the “policy and other contract liability requirements” is calculated, it becomes the numerator of a fraction, the denominator of which is the company’s investment earnings after expenses (in our example, $100 million/$150 million, or 66%). This fraction is the beneficiaries’ share— the percentage of investment income which, in the Code’s view, the company will use to satisfy its obligations to policyholders and other beneficiaries. In the end, the tax-free portion of investment income is determined by applying this fraction to each and every “item” of investment income, including tax-exempt income. See IRC § 804.

For present purposes, however, we need not take account of this fraction-related refinement (or of other refinements that we have glossed over). See generally 12 Mertens Law of Federal Income Taxation §§ 44A.19-40, at 68-179 (1989) (describing the 1959 taxing formula in detail); Pfeister & Pacer, Executive Guide to Federal Income Tax Planning for Life Insurance Companies (1981) (same). The main thing to keep in mind is that the Code uses the company’s average rate of return (15% in our example) to calculate the “pension plan” policyholders’ share of the company’s investment earnings, and it uses a special adjusted rate of return (related to the company’s average investment earnings) to calculate the “life insurance” policyholders’ share of the company’s investment earnings, but it uses the actual interest the company credits to accounts related to other kinds of business ($31 million in our example) to represent the related beneficiaries’ share.

This case primarily focuses upon the difference in the way the Code taxes “pension plan reserves” and other accounts. “Pension plan” treatment is sometimes (but only sometimes) better for a company than the “interest paid” treatment the Code provides in respect to other accounts. As a rule, the company will want to characterize an amount as a pension plan reserve if the rate at which it earned interest on its overall investments exceeds the rate at which the company actually credited interest on the amount. To see why this is so, recall that in our earlier example the company’s average earnings rate was 15%, and the amount of interest it actually credited on its $400 million of other retirement accounts was $31 million. If these $400 million of other accounts were instead deemed to be pension plan reserves, the company would be entitled to deduct 15% of them, or $60 million, rather than the $31 million it actually credited.

If, however, in a given year, the company earns interest on its total assets at a lower rate than it pays interest on its other retirement accounts, the company will prefer “interest paid” treatment to “pension plan” treatment. Suppose, for example, that interest rates rise suddenly, that the company must credit interest to its other accounts at rates approximating current rates, but that the company is heavily invested in long-term, lower interest securities; in that case, the company’s average earnings rate will be lower than the rate at which it actually credits interest to those accounts. In such circumstances, the com pany will want to classify accounts on which it paid interest at the going rate as neither “pension plan” nor “life insurance” reserves, for the company could then count as the beneficiaries’ share (and, in essence, deduct) the higher interest that it actually paid, rather than the lower amount that would result from multiplying the reserve by the company’s average earnings rate.

The circumstances described in the last paragraph, or analogous ones, appear to have led to the controversy before us. UNUM claims that the accounts held under its “deposit administration contracts” — accounts to which it credited a large amount of interest in 1978 and 1979 — are neither “life insurance reserves” nor “pension plan reserves,” and therefore deserve “interest paid” treatment, permitting UNUM (in effect) to deduct the interest it actually paid on these accounts.

2. Deposit Administration Contracts. Various companies and trustees (whom we shall call “Employers”) have entered into “deposit administration contracts” (which we shall refer to as “Contracts”) with UNUM to help create retirement programs for their employees. Under such contracts, the Employer typically deposits money into an account that UNUM maintains and which it calls a “Deposit Fund.” Within broad limits the Employer can decide how much to deposit and when to make deposits. UNUM pays interest on the account balances at no less than a guaranteed rate; this (retirement-related) interest accumulates tax free; and, eventually, the Employer withdraws some (or all) of the money, presumably when some, or all, of its employees reach retirement age. The contract permits the Employer to use the money that it withdraws to buy a lifetime annuity for the retiring employee. The employer might buy such an annuity from UNUM, it might (in effect) withdraw money for an employee and buy a similar annuity elsewhere, or it might simply withdraw money in a lump sum, give the money to the employee, and let the employee decide whether, or where, to buy an annuity.

3. Permanent rate guarantees. UNUM’s “deposit administration contracts” have an important special feature in respect to annuities. Attached to each contract is a table that lists the price of a one-dollar monthly annuity for a person aged fifty-five, fifty-six, and so on up to age seventy. UNUM guarantees these prices with respect to any annuity bought with money deposited by the Employer into its “Deposit Fund” during the first five years of the Contract’s life. That is, if an Employer uses money deposited during the first five years of the Contract’s life to buy annuities from UNUM, it can purchase them at the rates set forth in the table, regardless of when it actually buys the annuity. Consider, for example, an Employer who enters into a Contract with UNUM in 1970; the Employer can use money deposited through 1975 (plus interest earned by that money) to buy annuities at guaranteed rates set forth in the Contract, even if the Employer does not buy any annuities until, say, 1999, when its employees begin to retire. Because the Employer can take advantage of the Contract’s guaranteed annuity purchase rates at any time, even long after the five year period for depositing the money (to which the rates apply) has passed, the parties call the annuity price guarantees “permanent.” Of course, the Employer can also buy annuities with money deposited after the initial five-year period, but the Contract does not guarantee the rates that UNUM will charge for those annuities.

II

The Legal Question

The legal question before us is whether or not the accounts, or reserves, that UNUM maintained to satisfy its liabilities under the contracts just described amounted to a “pension plan reserve” within the meaning of the Code. See IRC § 805(d)(1). If not, the parties assume (and so, for the purposes of this case, do we) that UNUM may count as part of the tax-exempt beneficiaries’ share of investment income the full amount of interest that UNUM actually credited to the those accounts. See IRC § 805(a)(3) (tax-exempt portion of investment income includes “interest paid”); id. § 805(e)(1) (“interest paid” includes “interest on indebtedness”). If they amounted to a “pension plan reserve,” however, the parties (and we) assume that UNUM may not count the full amount of interest it actually credited as “interest paid.”

The Code provides a statutory answer to this question. It initially defines “pension plan reserves” as “that portion of the life insurance reserves” that meet certain requirements (which concededly are met here). See IRC § 805(d)(1). It then goes on to define the consequent key term “life insurance reserves” as “amounts”

(A) which are computed or estimated on the basis of recognized mortality or morbidity tables and assumed rates of interest, and

(B) which are set aside to ... liquidate ... future unaccrued claims arising from life insurance [or] annuity ... contracts ... involving, at the time with respect to which the reserve is computed, life ... contingencies....

provided that those “reserves” are

(C) required by law.

IRC § 801(b)(l, 2) (emphasis added). UNUM argues that it does not “compute[ ]” or “estimate[ ] on the basis of recognized mortality or morbidity tables” the amounts it must set aside to satisfy its “deposit administration contract” liabilities; hence, they are not “life insurance reserves;” hence they cannot be “pension plan reserves;” hence, they are entitled to “interest paid” treatment.

The district court fully understood why a life insurance company (or its regulators) would normally use “mortality or morbidity tables” to determine not only rates, but also the proper size of a reserve set aside to satisfy liabilities under, say, policies sold at those rates. The amount of a life insurance company’s future liabilities under the life insurance, or annuity, policies that it has sold, for example, depends, in part, upon how long the purchasers live. Once it sells a life insurance policy, then (depending on the policy’s terms) the longer the buyer lives, the more money the company may make (e.g., it may earn interest on premiums for a longer time); once it sells an annuity, the sooner the buyer dies, the fewer the annuity payments the company must make. While the company cannot know how long an individual buyer will live in fact, the company, by using recognized mortality or morbidity tables, can know how long groups of persons with similar characteristics will live on average; and it can thereby know whether or not the reserves it sets aside should prove sufficient to cover the payments it will have to make to similar groups of its own buyers. Before it sells life insurance or annuities, the company will likely consult such tables to determine rates; after it sells the policies, it will likely check such tables from time to time to see whether mortality estimates have changed, making its reserves either overly abundant or insufficient. Regulators, too, will use such tables for related purposes.

The district court also recognized that the annuity purchase rate guarantees in UNUM’s “deposit administration contracts” create a mortality-based financial risk. The risk arises because, when UNUM sets these guaranteed rates, it uses mortality assumptions contained in mortality tables, and these mortality assumptions could be obsolete by the time employers actually buy annuities for their employees. Suppose, for example, UNUM promises in a Contract that it will sell an annuity (purchased with money deposited during the first five years) to a sixty-five year-old at a price of $100 per dollar of monthly annuity benefit (i.e., the annuity buyer can pay $1000 and thereafter receive $120 per year for the rest of his life). Now suppose that, long before the annuity is actually purchased, a new treatment for heart disease causes the life expectancy of sixty-five year-olds to increase dramatically. If the new average lifespan of a sixty-five year-old exceeds the expected lifespan appearing in the table that UNUM used when it calculated its guaranteed rates, UNUM could wind up losing money if Employers buy annuities for retiring sixty-five year-old employees at the guaranteed rate. Although UNUM could minimize this risk by using extremely “conservative” mortality assumptions when it sets its rate guaran tees, it could not, at least in theory, entirely eliminate it, because people could always wind up living longer on average than UNUM initially expected.

Nonetheless, despite the theoretical “mortality based” risk, and the theoretical consequent possibility that UNUM would consult mortality tables to determine the adequacy of its contract-related reserves, the district court, after hearing factual and expert evidence, held that UNUM did not in fact consult any “recognized mortality or morbidity tables” in determining the size of those reserves. It held that the reserves, therefore, failed to satisfy statutory condition “(A),” see IRC § 801(b)(1)(A), that they were not “life insurance reserves,” and that they therefore could not be “pension plan reserves.”

We have reviewed the record to determine whether the district court lawfully could reach this conclusion. We believe that the record provides adequate support.

Ill

The Evidence

The evidence at trial was that UNUM listed its liabilities under its “deposit administration contracts” on its Annual Statement under a subheading labelled “Deposit Administration Funds” (though under a heading labelled “Annuities.”) Its witnesses said that UNUM calculated the amount of this liability in the same way a bank might compute its liability to a passbook holder: it added together all Employer deposits and interest payments on those deposits, and then it subtracted all withdrawals, withdrawal penalties, and administrative charges. This amount, which represented the total amount of money that Employers could withdraw from their respective accounts, was the figure that UNUM listed as its total liability under the “Deposit Administration Funds.” Elsewhere, UNUM listed its liabilities for annuities that Employers actually bought with these funds (after the Employers bought them); but the parties agree that these amounts are not entitled to “interest paid” treatment; and they are not here in issue. Given the evidence, the record amply supports the proposition that neither the amount UNUM listed as “Deposit Administration Funds,” nor any other amount representing liability under UNUM’s “deposit administration contracts” (before the Employer actually bought an annuity), was actually “computed or estimated on the basis of recognized mortality or morbidity tables,” IRC § 801(b)(1)(A), particularly in light of the following features.

First, UNUM’s expert made it particularly plausible to believe that UNUM did not use mortality tables when he explained why, despite the fact that money deposited by Employers could be used to purchase annuities, UNUM, in fact, did not calculate its liability under its “deposit administration contracts” in the way that companies normally calculate their liability under annuity contracts. As we explained earlier, when a person actually buys an annuity, the company can only estimate its liability by looking up the annuitant’s estimated lifespan on a mortality table. When an Employer with a “deposit administration contract” deposits money into its account, however, that money is not immediately applied to purchase annuities. Indeed, UNUM for the most part has no idea whether that money will ever be used to purchase an annuity, or whether it will simply be withdrawn in a lump sum. Furthermore, even if UNUM knew that the money would one day be used to purchase an annuity, UNUM for the most part had no way of knowing ahead of time mortality-related characteristics (e.g., age, smoking habits) the individual annuitant might have; hence, it could not easily have looked up an individual estimated lifespan on a mortality table even if it had wanted to do so; hence, it preferred to wait until the Employer chose to buy annuities for individuals and it could know the characteristics of the annuitants.

Second, the evidence to the contrary consisted only of speculation by experts about what UNUM might have done, or could have done, not about what UNUM actually did. For example, one expert said that after UNUM calculated each of its liabili ties “item by item,” UNUM then determined whether the sum of these liabilities was an “adequate statement of the company’s total obligation^]. ” Tr. at 113 (testimony of UNUM’s expert actuary). He also said that UNUM would consider, as part of this overall “adequacy” determination, the risk that the guaranteed purchase rates in its “deposit administration contracts” would no longer be profitable if and when they were actually exercised. But this risk of unprofitability depended on a (potentially infinite) variety of factors. While one factor was the possibility that individuals who purchased annuities 5, 10 or 20 years hence could wind up living longer than UNUM initially expected, other factors could also affect the eventual profitability of a guaranteed rate, factors such as potentially increased regulation, higher taxes, increased administrative costs, and so on. Moreover, the experts pointed out that the mortality-related risk may have had little practical impact on the adequacy of UNUM’s total reserves, for a life expectancy increase that would have diminished annuity profitability might simultaneously have increased life insurance profitability. The great variety of the many risks, and the potential insignificance of the mortality-based risk, makes it impossible to jump from the premise (1) that UNUM, when determining whether its overall statement of liabilities was “adequate,” might have tried to account for the risk that an unknown number of employees purchasing annuities at some unknown future date would live longer than expected, to the conclusion (2) that UNUM actually did so.

Third, the likelihood that UNUM did not, in any meaningful sense, take account of the mortality risk in calculating the size of its likely future “deposit administration contract” liability is reinforced by the record evidence indicating that Maine’s insurance rules did not require it to do so. As far as the record reveals, Maine did not legally require UNUM to calculate its “deposit administration contract” liability through reference to mortality tables: Maine regulators are fully aware of the risks created by guaranteed annuity purchase rates; and, with respect to some contracts, Maine requires insurance companies to account for this risk even before the annuity is purchased. For example, if an individual opens an account that enables him to make deposits, earn tax-free interest, and then use the balance at a designated date to purchase an annuity at a guaranteed rate, (the record suggests) Maine regulators require the life insurance company to calculate its liability for that account using a recognized mortality table; the company must assume that the current balance, increased by interest, will be used on the designated date to buy an annuity, and it must calculate its liability by looking up the person’s relevant mortality-related characteristics (as they will be at the time of purchase) on relevant mortality tables. According to the UNUM employee who actually calculated UNUM’s “deposit administration contract” liability, however, Maine regulators, while fully aware that UNUM (in respect to those contracts) did not refer to mortality tables until the Employer actually bought an annuity, nonetheless never “suggested to UNUM that UNUM’s computations [with respect to this liability were] required to take into account assumptions of mortality.” Trial Transcript at 192 (Testimony of Frederick Yosua, an UNUM vice-president and corporate actuary). Given that the relevant mortality-based risk was highly speculative and the calculations difficult to make or not very accurate, we cannot say that Maine’s regulators’ views, as indicated in this record, are unreasonable or could not reflect Maine’s law.

Fourth, nothing in the record suggested that UNUM changed its mortality-table-eonsultation procedures from year to year, or otherwise deliberately avoided any actual reference to those tables as a way to avoid, or to change, its tax liability.

Given this state of the record, we cannot find any plausible sense in which one might say, in the statutory words, that UNUM “computed or estimated” its liability for its “deposit administration contracts” on “the basis of recognized mortality or morbidity tables.” IRC § 801(b)(1)(A).

IV

The Government’s Argument

The government argues that, because the purchase rate guarantees in UNUM’s “deposit administration contracts” create a mortality risk, and because UNUM calculated those rates using recognized mortality tables, that UNUM must also have computed its liabilities under these contracts using mortality tables. It summarizes its argument in its brief as follows:

[T]he insurer [under a deposit administration contract with permanent annuity purchase rate guarantees] is obligated to accumulate contributions received from the contract holder at a guaranteed rate of interest and with that accumulation to provide annuities to the contract holders’ employees at rates guaranteed by the contract. That obligation creates a liability to provide annuities at guaranteed rates with respect to every dollar received from the contract holders. The amount of that liability depends, of course, on the probable rate of mortality among employees after retirement, as well as on the anticipated interest to be earned both before and after retirement. Manifestly then, every dollar paid to an insurance company under a deposit administration contract creates a liability whose amount can be computed only by the use of mortality and interest tables.

Appellant’s Brief at 18-19 (emphasis added). We have tried to understand how the government could jump from its premise, which we have underlined, to the necessary conclusion — that UNUM must have computed its liability under the Contracts using mortality tables. To do so, we think the government must convince us to accept one of three possible arguments, all variations on the same theme.

First, the government might mean that the Deposit Funds liability is a § 801(b)(1) “life insurance reserve” simply because the guaranteed purchase rates in the Contracts create a risk of loss that cannot be measured precisely without considering mortality tables. As we have said, UNUM, in time, runs the risk that Employers will choose an annuity for some of their employees, that they will invoke the guaranteed rate when they do so, and that UNUM will lose money because its guaranteed rates turn out to be too low. To measure this risk precisely, UNUM would have to check, among other things, the adequacy of the mortality tables on which its guaranteed rates are based.

It cannot be, however, that the mere existence of a risk, the precise measurement of which requires one to look at mortality tables, can make the amount giving rise to the risk a “life insurance reserve” within the meaning of the Code. For one thing, that is not what the Code says. The relevant provision, IRC § 801(b)(1)(A) defines “life insurance reserves” solely in terms of how a company, in fact, computes a particular “amount.” It refers to “amounts ... which are computed or estimated on the basis of recognized mortality ... tables.” IRC § 801(b)(1). It does not speak of “amounts” that “might be,” or, “in principle, could be,” or, “for greatest accuracy ought to be,” computed in that way.

For another thing, to call an amount a “life insurance reserve” simply because it creates a certain kind of mortality-based risk of loss would prove far too much. Any contract, for example, that permits depositors to buy annuities at rates guaranteed for any period of time creates some risk of mortality-based loss for the offering company. Even if the rate guarantee is not “permanent,” but “temporary,” requiring the depositor to exercise the option within the next five years (or five months or five days), a company selling such a contract would run the risk that, say, the invention of a life-prolonging miracle cure during that time would dramatically increase its contract liability, a liability that it could measure precisely only by (1) estimating how many persons will choose the annuity purchase option, and (2) comparing the mortality tables built into the guaranteed rates with current mortality expectations. The government, however, has specifically ruled that a deposit administration contract containing the type of “tempo rary” rate guarantee that we have just described is not a “life insurance reserve,” even though such a contract creates the same kinds of risk as a contract containing a “permanent” rate. See Rev.Rul. 77-286, 1977-2 Cum.Bull. 228. (The relative magnitude of the risks would seem to depend, in large part, on the comparative length of the guarantee and the rates involved, not simply on whether the rate guarantees are “temporary” or “permanent”). The government, by this Ruling, apparently (and in our view, properly) concedes that the mere existence of a risk of the type we have described is not, by itself, sufficient to satisfy the requirements of IRC § 801(b)(1). It therefore is not surprising that the language of the statute focuses on the way an amount is, in fact, calculated, rather than the risks associated with that amount.

Second, the government seems to argue at times as if the fact (1) that UNUM likely referred to mortality when it set its guaranteed rates, shows (2) that UNUM referred to mortality tables when it determined the adequacy of its reserves to satisfy its obligations. The problem with this argument is that the conclusion does not follow from the premise. A company might use a mortality table to set a rate in respect to an offer while using some other basis for creating a reserve for liabilities incurred to one who accepts; or the company might decide to create no reserve at all until it learns who accepted; and the record here indicates that UNUM (likely with Maine’s approval) followed the second path.

Third, the government sometimes argues as if (1) the fact that Maine requires UNUM to state its overall liabilities in a way that accurately reflects its likely future obligations (and also shows enough assets to cover them), and (2) the fact that UNUM could not state its precise liabilities under the relevant Contracts without looking at mortality tables, together show (3) that UNUM did consult mortality tables when it stated its liabilities. The problem with this variation on the argument again lies in the fact that the conclusion does not follow from the premise. The statute says that “amounts” are not a § 801(b)(1) “life insurance reserve” unless they “are calculated on the basis of recognized mortality or morbidity tables.” IRC § 801(b)(1)(A) (emphasis added). The record does not show that UNUM, in fact, did calculate the amount of its reserves by using a “recognized mortality table.” Of equal importance, the record indicates that UNUM’s calculation methods were reasonable, that Maine’s insurance regulators did not require more, and that UNUM did not decide what, or what not to consult purely for tax reasons. Thus, we need not decide whether there are actions on UNUM’s part, or requirements of Maine law, that could lead a court to treat UNUM as if, in fact, it implicitly or explicitly calculated its reserves by using a mortality table, even though, in reality, it did not. We need, and do, only hold that, on this record, the facts, as the district court found them, about what UNUM in fact did (or did not do) are adequate to support its legal conclusion that the “amounts” in question were not calculated on the basis of “recognized mortality ... tables.”

V

Legal Authority

The case law offers strong support for our holding that, at least ordinarily, an amount does not satisfy the “life insurance reserve” definition in IRC § 801(b)(1) unless the company actually calculates its liabilities using mortality tables, irrespective of the existence of a mortality risk. In Union Mutual Life Ins. Co. v. United States, 570 F.2d 382, 396-97 (1st Cir.), cert. denied, 439 U.S. 821, 99 S.Ct. 87, 58 L.Ed.2d 113 (1978), for example, a life insurance company stated an additional liability that arose from an offer permitting policyholders (for an added fee) to buy more life insurance without a further medical examination. The liability created by this offer was, of course, mortality-based, and the company measured its size using a recognized mortality table. As part of this calculation, however, the company, without justification, assumed that all of its policyholders would exercise the “added insur- anee” option; consequently, in this court’s view, the company had not used mortality tables in a meaningful way. For that reason, the court held that the additional liability established by the company was not “ ‘calculated or estimated on the basis of mortality or morbidity tables.’ ” Union Mutual, 570 F.2d at 396 (quoting IRC § 801(b)(1)). Our result here, where the company not only did not use the tables in a meaningful way, but rather did not use them at all, would seem to follow a fortio-ri.

In Group Life Ins. v. United States, 660 F.2d 1042 (5th Cir.1981), cert. denied, 457 U.S. 1132, 102 S.Ct. 2958, 73 L.Ed.2d 1349 (1982), the Fifth Circuit considered a provision in a life insurance contract that permitted a policyholder who became disabled for more than six months to keep his life insurance in effect without paying any premiums. Because the risk of death is greater for a disabled, than a nondisabled, person, the Company created an additional reserve to account for this additional obligation. Although the company specifically created the reserve to reflect a special mortality risk, the Fifth Circuit held that it did not qualify as a “life insurance reserve” because the Company, on the advice of an industry report, calculated the reserve simply by taking 75% of the face value of the policy, rather than by referring to a “recognized mortality table.” See Group Life, 660 F.2d at 1054 (finding “no proof” that “this 75% figure ... was derived from [recognized] mortality tables”); cf. Aetna Life Ins. Co. v. United States, 16 Cl.Ct. 364 (1989) (holding that a reserve created for exactly the same purpose as the one in Group Life, but calculated on the basis of a recognized mortality table, was a life insurance reserve); Commissioner v. Monarch Life Ins. Co., 114 F.2d 314, 320 (1st Cir.1940) (same); Rev.Rul. 70-190, 197-1 C.B. 150. Again, our result here, where the Company not only did not, but also had no legal obligation to, consider mortality tables, would seem to follow a fortiori.

The government cites two cases in which courts held that reserves for an option to purchase an annuity at guaranteed prices were “life insurance reserves.” Lincoln Nat’l Life Ins. Co. v. United States, 582 F.2d 579, 217 Ct.Cl. 515 (1978); Equitable Life Ins. Co. v. Commissioner, 73 T.C. 447 (1979). In each of those cases, however, the company actually used “recognized mortality tables” to calculate the reserve. Indeed, the companies initially established the exact same reserve that they would eventually become obligated to establish once annuities were actually purchased (only discounted to reflect the fact that the annuities would be purchased in the future). These cases are different from this case, for here, UNUM did not, and was not legally obligated to, establish such a reserve.

Taken together, these cases stand for the proposition that, at least ordinarily, it is what a company actually does, not what it theoretically might do, that determines whether IRC § 801(b) applies. That provision, by its terms, normally requires the company actually to use “recognized mortality tables,” rather than some other system, for measuring its liability under a contract that creates a mortality risk. The IRS’s own rulings reflect the importance of this distinction. Compare Rev.Rul. 68-185,1968-1 C.B. 317 (holding that a reserve for credit life insurance policies was a life insurance reserve because it was calculated using mortality tables and assumed interest rates) with Rev.Rul. 69-302, 1969-1 C.B. 186 (holding that a reserve for credit life insurance policies was not a life insurance reserve because it was not calculated using mortality tables and assumed interest rates). To find for the government here would require a departure from rather well-established legal interpretation.

YI

The Treasury Regulation

The Government, for the first time on this appeal, points to a Treasury Regulation, Treas.Reg. § 1.801-4(d) (1985), as support for its position. That Regulation lists examples of reserves that normally qualify as “life insurance reserves.” It includes reserves held under “deposit administration contracts.” The Regulation states:

The following reserves, provided they meet the requirements of section 801(b) ... are illustrative of reserves which shall be included as life insurance reserves:

(5) Reserves held under deposit administration contracts. Generally, the reserves held by a life insurance company on both the active and retired lives under deposit administration contracts will meet the requirements of section 801(b)....

(Emphasis added.) The government also refers to a subsequent General Counsel Memorandum that describes the history of this regulation; the Memorandum says that Treasury wrote these words (particularly, the words “reserves ... on ... active ... lives”) only after insurance industry representatives pointed out that the typical deposit administration contract obligates the insurer to provide annuities that (in respect to some of the deposited money) it promises to sell at rates computed well in advance and on the basis of recognized mortality tables and assumed interest rates. See Gen.Couns.Mem. 38,378 (May 16, 1980). The government contends that the contracts at issue here are the same as those that motivated the Treasury Regulation, that they therefore fall within the word “generally,” and that, in light of the Regulation, we must decide in its favor. See Chevron, Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 844, 104 S.Ct. 2778, 2782, 81 L.Ed.2d 694 (1983) (“[LJegislative regulations must be given controlling weight unless they are arbitrary, capricious, or manifestly contrary to the statute.”); Commissioner v. South Texas Lumber Co., 333 U.S. 496, 501, 68 S.Ct. 695, 699, 92 L.Ed. 831 (1948) (“This Court has many times declared that Treasury regulations must be sustained unless unreasonable and plainly inconsistent with the revenue statutes.”); Udall v. Tallman, 380 U.S. 1, 16, 85 S.Ct. 792, 801, 13 L.Ed.2d 616 (1965) (“When the construction of an administrative regulation rather than a statute is in issue, deference is even more clearly in order.”).

In our view, however, the Regulation cannot make the difference for which the government hopes. Read literally, it does not compel a holding that the reserves in question qualify as “life insurance reserves.” It says that its listed examples qualify only if they “meet the requirements of section 801(b).” It adds that “deposit administration contracts” will “generally” meet those requirements, but it does not specify what it means by “generally;” that is, it does not specify what it is about such contracts that brings them outside any exception and within the rule.

The government tries to make up for this deficiency in arguing that by “generally” the regulation means to embrace contracts with “permanent” annuity rate guarantees {i.e., rate guarantees applicable to all money deposited within, say, the contract’s first five years, irrespective of when the Employer actually purchases an annuity with that money), and to exclude similar contracts that have only “temporary” annuity rate guarantees {i.e., guarantees applicable only to annuities that the Employer chooses to buy within, say, five years from the time it enters into the contract). The government adds that the insurance industry itself convinced it to add “deposit administration contracts” to the Regulation’s list precisely because such contracts typically contained annuity options with “permanent rate guarantees.” See Gen.Couns. Mem. 38,378 at 5-6.

The government’s explanation does not go far enough, however, for it does not explain why the Regulation would distinguish between contracts with different purchase rate guarantees. The explanation cannot lie in the simple existence of a mortality risk, for both types of guarantee create such risks. Rather, given the statutory language, the explanation would seem most probably to lie in the fact that the insurance industry more likely actually uses mortality tables to calculate the liability created by mortality risk when (quite possibly longer) “permanent,” rather than (quite possibly shorter) “temporary,” rate guarantees are at stake. Although we are, of course, speculating, the government of fers us no other explanation; and if we are right about the reason, UNUM’s case would, of course, more appropriately fall outside the scope of the Regulation’s word “generally.”

We need not decide this point definitively, however. Nor need we decide such related matters as the degree of legal respect we must show an agency interpretation that we cannot fully understand. Cf. Bowles v. Seminole Rock Co., 325 U.S. 410 at 413-14, 65 S.Ct. 1215, at 1217, 89 L.Ed. 1700 (1945) (an agency’s construction of its own regulations deserves “controlling weight unless it is plainly erroneous or inconsistent with the regulation”). That is because the statute speaks specifically of “amounts” that “are” calculated on the basis of “recognized mortality or morbidity tables.” See IRC § 801(b)(1). A regulation that brought within this definition a reserve that plainly was not calculated on the basis of such tables, in the absence of some special reason that could convince a court to presume that it was so calculated, would fall outside the scope of Treasury’s statutory regulation-writing authority. We shall not permit the Treasury to read its regulation in a manner that would make it unlawful.

The government adds, and emphasizes, that the Congress that wrote this language intended it to provide favorable tax treatment to insurance companies managing qualified pension plan funds for small employers, see S.Rep. No. 291, 86th Cong., 2d Sess. at 8, 1959-2 Cum.Bull. 770, 775, and that the “ ‘most common vehicle used by life insurance companies for qualified [pension] plans is the Deposit Administration Contract.’ ” Reply Brief for Appellant at 6 (quoting Van Mieghem & Brown, Federal Income Taxation of Insurance Companies 239 (1986)). We do not see how these observations can make a significant difference, however, given the government’s failure to explain how our holding would deny favorable tax treatment to companies administering to the needs of small employers; nor do we see how the government can overcome the statute’s language. Thus, in our view, the more plausible, and the only legally valid, reading of the Regulation is that it excludes from its term generally an amount stated as liability under a “deposit administration contract” when the company does not (in any plausible sense) use a recognized mortality table to calculate the amount.

The district court entered judgment for UNUM and ordered the government to refund $3,301,106 of tax and assessed interest that the government had previously collected. The government says that this figure overstates the amount of interest to which UNUM is entitled by $74,224.03. UNUM has stated to us that it “stands ready to resolve [this] difference in interest calculations” with the government. Brief for the Appellee at 33. We therefore affirm the judgment of the district court with respect to all but the $74,224.03 that remains in dispute and vacate the judgment with respect to that $74,244.03. On remand, the court should determine what portion, if any, of that $74,244.03 should be awarded to UNUM.

Affirmed in part, vacated in part, and remanded for the further proceedings that we have ordered.