EASTERBROOK, Circuit Judge.
Several subsidiaries of Sears, Roebuck & Co. sell insurance. One, Allstate Insurance CO., underwrote some of the risks of the parent corporation. Two others wrote mortgage insurance, promising to pay lenders if borrowers defaulted. Because Sears and all other members of the corporate group file a consolidated tax return, disputes about the tax consequences of these transactions affect the taxes of the entire [860]*860group. The Commissioner of Internal Revenue assessed the, group with deficiencies exceeding $2.5 million for the tax years 1980-82. Whether the group owes this money depends on the proper characterization of the two kinds of transaction.
An insurer may deduct from its gross income an amount established as a reserve for losses. 26 U.S.C. § 832. Until 1986 it could deduct the entire reserve; today it must discount this reserve in recognition of the fact that a dollar payable tomorrow is worth less than a dollar today. Tax Reform Act of 1986 § 1023, 100 Stat. 2085, 2399 (1986). These transactions occurred before 1986, and in any event we deal with the existence rather than the size of the deduction. Allstate created and deducted reserves to cover casualties on policies it issued to Sears. The Commissioner disallowed these deductions (and made some related adjustments), reasoning that the shuffling of money from one corporate pocket to another cannot be “insurance.” The Tax Court disagreed. It distinguished captive subsidiaries (which write policies for the parent corporation but few or no others) from bona fide insurance companies that deal with their corporate parents or siblings at market terms. 96 T.C. 61 (1991).
The two subsidiaries underwriting mortgage insurance estimated losses as of the time the underlying loans went into default. The Commissioner contended that these insurers could not establish deductible loss reserves until the lenders obtained good title to the mortgaged property, because the insurance policies made a tender of title a condition precedent to the insurers’ obligation to pay. The Tax Court agreed with this conclusion, rejecting the insurers’ argument that the Internal Revenue Code permits them to deduct loss reserves required by state law, as these reserves were.
The judges of the Tax Court split four ways. Judges Korner, Shields, Hamblen, Swift, Gerber, Wright, Parr, Colvin, and Halpern joined Judge Cohen’s opinion for the majority. Judges Chabot and Parker would have ruled for the Commissioner on both issues; Chief Judge Nims and Judge Jacobs would have ruled for Sears on both issues. Judge Whalen concluded that the majority had things backward: that Sears should have prevailed on the mortgage insurance issue but lost on the subsidiary issue. We join Chief Judge Nims and Judge Jacobs.
I
Allstate is a substantial underwriter, collecting more than $5 billion in premiums annually and possessing more than $2 billion in capital surplus. During the years at issue, Allstate charged Sears approximately $14 million per year for several kinds of insurance. Some 99.75% of Allstate’s premiums came from customers other than Sears, which places 10% to 15% of its insurance with Allstate. The Commissioner’s brief concedes that “[pjolicies issued to Sears by Allstate were comparable to policies issued to unrelated insureds. With respect to the execution, modification, performance and renewal of all of the policies in issue, Allstate and Sears observed formalities similar to those followed with respect to the insurance policies issued by Allstate to unrelated insureds. In addition, the premium rates charged by Allstate to Sears were determined by means of the same underwriting principles and procedures that were used in determining the premium rates charged to unrelated insureds, and were the equivalent of arm’s-length rates.” The Tax Court made similar findings, although not nearly so concisely.
Allstate, founded in 1931, has been selling insurance to Sears since 1945. Everyone, including the Commissioner, has taken Allstate as the prototypical non-captive insurance subsidiary. Until 1977 the Internal Revenue Service respected transactions between non-captive insurers and their parents. In that year the Commissioner decided that a wholly owned subsidiary cannot “insure” its parent’s operations, even if the subsidiary’s policies are identical in terms and price to those available from third parties. Rev.Rul. 77-316,1977-2 C.B. 53. Examples given in this revenue ruling all dealt with captives that had no customers [861]*861outside the corporate family. After issuing the ruling the Service continued to believe that subsidiaries engaged in “solicitation and acceptance of substantial outside risks” could provide insurance to their parents. G.C.M. 38136 (Oct. 12, 1979). But in 1984 the General Counsel reversed course, G.C.M. 39247 (June 27, 1984), and the Commissioner later announced that all wholly owned insurance subsidiaries should be treated alike. Rev.Rul. 88-72, 1988-2 C.B. 31, clarified, Rev.Rul. 89-61, 1989-1 C.B. 75. Our task is to decide whether this is correct. We therefore disregard details, which may be found in the Tax Court’s opinion. Like the Commissioner, we deem immaterial the nature of the risks Allstate accepted, the terms the parties negotiated, and the precise deductions taken.
If Sears did no more than set up a reserve for losses, it could not deduct this reserve from income. United States v. General Dynamics Comp., 481 U.S. 239, 107 S.Ct. 1732, 95 L.Ed.2d 226 (1987). Firms other than insurance companies may deduct business expenses only when paid or accrued; a reserve is deductible under § 832 only if the taxpayer issued “insurance.” “Self-insurance” is just a name for the lack of insurance — for bearing risks oneself. According to the Commissioner, “insurance” from a subsidiary is self-insurance by another name. Moving funds from one pocket to another does nothing, even if the pocket is separately incorporated. If Subsidiary pays out a dollar, Parent loses the same dollar. Nothing depends on whether Subsidiary has other customers; there is still a one-to-one correspondence between its payments and Parent’s wealth. So although Allstate may engage in the pooling of risks, and thus write insurance, Sears did not purchase the shifting of risks, and thus did not buy insurance. Unless the transaction is insurance from both sides — unless it “involves risk-shifting [from the client’s perspective] and risk-distributing [from the underwriter’s]”, Helvering v. Le Gierse, 312 U.S. 531, 539, 61 S.Ct. 646, 649, 85 L.Ed. 996 (1941)—it is not insurance for purposes of the Internal Revenue Code. The Commissioner asks us to pool the corporate family’s assets to decide whether risk has been shifted. This is .the “economic family” approach of Rev.Rul. 77-316, which the Service sometimes supplements with a “balance sheet” inquiry under which a transaction is not insurance if it shows up on both sides of a corporation’s balance sheet.
No judge of the Tax Court has ever embraced the IRS’s “economic family” approach, which is hard to reconcile with the doctrine that tax law respects corporate forms. Molien Properties, Inc. v. CIR, 319 U.S. 436, 63 S.Ct. 1132, 87 L.Ed. 1499 (1943). Although the Commissioner may recharacterize intra-corporate transactions that lack substance independent of their tax effects, cf. Gregory v. Helvering, 293 U.S. 465, 55 S.Ct. 266, 79 L.Ed. 596 (1935); Yosha v. CIR, 861 F.2d 494 (7th Cir.1988)—which supports disregarding captive insurance subsidiaries — the “economic family” approach asserts that all transactions among members of a corporate group must be disregarded. Even the ninth circuit, which in citing Rev.Rul. 77-316 favorably has come the closest to the Commissioner’s position, has drawn back by implying that subsidiaries doing substantial outside business cannot be lumped with true captives into a single pot. Carnation Co. v. CIR, 640 F.2d 1010 (9th Cir.1981); Clougherty Packing Co. v. CIR, 811 F.2d 1297, 1298 n. 1 (9th Cir.1987).
What is “insurance” for tax purposes? The Code lacks a definition. Le Gierse mentions the combination of risk shifting and risk distribution, but it is a blunder to treat a phrase in an opinion as if it were statutory language. Zenith Radio Corp. v. United States, 437 U.S. 443, 460-62, 98 S.Ct. 2441, 2450-51, 57 L.Ed.2d 337 (1978). Cf. United States v. Consumer Life Insurance Co., 430 U.S. 725, 740-41, 97 S.Ct. 1440, 1448-49, 52 L.Ed.2d 4 (1977). The Court was not writing a definition for all seasons and had no reason to, as the holding of Le Gierse is only that paying the “underwriter” more than it promises to return in the event of a casualty is not insurance by any standard. Life insurance passes outside a decedent’s estate, making it advantageous to turn (taxable) assets of [862]*862the estate into insurance proceeds. Less than a month before her death, an elderly woman bought a policy denominated life insurance. .The policy named a death benefit of $25,000 and carried a premium of $23,000. As part of the package, the “insurer” required the beneficiary to buy an annuity contract for $4,000. If the beneficiary died immediately, the insurer was $2,000 to the good; if she lived, the premiums were more than enough to fund the promised annuity payment and death benefit. So no risks were being spread, transferred, pooled, whatever. As the Court observed, there was no insurance risk; the buyer of the policy expected to die soon, and the issuer expected to turn the proceeds over to the heirs, keeping an administrative fee for the service of removing the assets from the estate. Le Gierse, like Gregory and Yosha, shows that substance prevails over empty forms. Sears, by contrast, had insurable risks. The Commissioner does not deny that if Sears had purchased from Hartford or Aetna the same policies it purchased from Allstate, these would have been genuine “insurance.” Forms there were, but not empty ones — and taxes usually depend on form, as the Commissioner trumpets whenever this enlarges the revenue. E.g., Howell v. United States, 775 F.2d 887 (7th Cir.1985). Distinctions with little meaning to the populace — for example, income at 11:59 p.m. on December 31 versus income at 12:01 a.m. on January 1, or wages plus the promise of a pension versus higher wages used to purchase an annuity — produce large differences in tax.
Doubtless a casualty that leads Allstate to reimburse Sears does not bring cash into the corporate treasury the same way a payment from Hartford would. A favorable loss experience for Sears cuts Allstate’s costs, and thus augments the group’s aggregate wealth, by more than the same reduction in losses would produce if Hartford issued the policy. Yet the Commissioner does not push this as far as he could. Corporate liability is limited by corporate assets. Corporations accordingly do not insure to protect their wealth and future income, as natural persons do, or to provide income replacement or a substitute for bequests to their heirs (which is why natural persons buy life insurance). Investors can “insure” against large risks in one line of business more cheaply than do corporations, without the moral hazard and adverse selection and loading costs: they diversify their portfolios of stock. Instead corporations insure to spread the costs of casualties over time. Bad experience concentrated in a single year, which might cause bankruptcy (and its associated transactions costs), can be paid for over several years. See generally David Mayers & Clifford W. Smith, Jr., On the Corporate Demand for Insurance, 55 J. Bus. 281 (1982). Much insurance sold to corporations is experience-rated. An insurer sets a price based on that firm’s recent and predicted losses, plus a loading and administrative charge. Sometimes the policy is retrospectively rated, meaning that the final price is set after the casualties have occurred. Retrospective policies have minimum and maximum premiums, so the buyer does not bear all of the risk, but the upper and lower bounds are set so that almost all of the time the insured firm pays the full costs of the losses it generates. Both experience rating and retrospective rating attempt to charge the firm the full cost of its own risks over the long run, a run as short as one year with retrospective rating. The client buys some time-shifting (very little in the case of retrospective rating) and a good deal of administration. Insurers are experts at evaluating losses, settling with (or litigating against) injured persons, and so on. A corporation thus buys loss-evaluation and loss-administration services, at which insurers have a comparative advantage, more than it buys loss distribution. If retrospectively rated policies, called “insurance” by both issuers and regulators, are insurance for tax purposes — and the Commissioner’s lawyer conceded for purposes of this case that they are — then it is impossible to see how risk shifting can be a sine qua non of “insurance.”
The Commissioner insists that “shifting risk to third-party insureds” is an essential ingredient of insurance, but what does this mean? Take term life insurance. One [863]*863thousand persons at age 30 pay $450 each for a one-year policy with a death benefit of $200,000. In a normal year two of these persons will die, so the insurer expects to receive $450,000 and disburse $400,000. Of course more may die in a given year than the actuarial tables predict. But as the size of the pool increases the law of large numbers takes over, and the ratio of actual to expected loss converges on one. The absolute size of the expected variance increases, but the ratio decreases.
Risk-averse buyers of insurance shuck risk. Risk-neutral insurers match risks. No third party gets extra risk. Each person’s chance of dying is unaffected; the financial consequences of death are shared. Joseph E. Stiglitz, professor of economics at Stanford, one of the leading students of risk and insurance, and an expert witness for Sears, put things nicely in saying that insurance does not shift risk so much as the pooling transforms and diminishes risk. See Richard A. Posner, Economic Analysis of Law 103 (4th ed. 1992). Insurance companies, with diversified investors and oodles of potential claims, are effectively risk-neutral. So everyone gains. The insureds willingly pay the loading charge to reduce their financial variance. The investors in the underwriters make a profit.
Convergence through pooling is an important aspect of insurance. Allstate puts Sears’s risks in a larger pool, performing one of the standard insurance functions in a way that a captive insurer does not. More: Allstate furnishes Sears with the same hedging and administration services it furnishes to all other customers. It establishes reserves, pays state taxes, participates in state risk-sharing pools (for insolvent insurers), and so on, just as it would if Sears were an unrelated company. States recognize the transaction as “real” insurance for purposes of mandatory-insurance laws (several of the policies were purchased to comply with such laws for Sears’s auto fleet, and for workers’ compensation in Texas). From Allstate’s perspective this is real insurance in every way. It must maintain the reserves required by state law (not to mention prudent management). Sears cannot withdraw these reserves on whim, and events that affect their size for good or. ill therefore do not translate directly to Sears’s balance sheet. It therefore does not surprise us that the Tax Court, while accepting the Commissioner’s view that true captives do not write insurance, believes that insurance affiliates with substantial business from outside the group are genuine insurers. E.g., Gulf Oil Corp. v. CIR, 89 T.C. 1010, 1025-27 (1987) (dictum), aff’d in relevant part, 914 F.2d 396 (3d Cir.1990); AMERCO v. CIR, 96 T.C. 18 (1991) (52% to 74% writing for unrelated parties); Harper Group v. CIR, 96 T.C. 45 (1991) (30% writing for unrelated parties). So, too, courts of appeals have allowed the Commissioner to recharacterize “captive” cases as self-insurance without extending this principle to firms with substantial outside business. Beech Aircraft Corp. v. United States, 797 F.2d 920 (10th Cir.1986); Stearns-Roger Corp. v. United States, 774 F.2d 414 (10th Cir.1985). One court has held that fraternal corporations may write genuine “insurance” for each other, although they do no business outside the corporate group. Humana Inc. v. CIR, 881 F.2d 247 (6th Cir.1989).
Power to recharacterize transactions that lack economic substance is no warrant to disregard both form and substance in the bulk of cases. The Tax Court has given up the effort to find a formula, instead listing criteria such as insurance risk, risk shifting, risk distribution, and presence of forms commonly accepted as insurance in the trade. 96 T.C. at 99-101 (this case); Harper, 96 T.C. at 57-58 (opinion by Judge Jacobs describing this as a “facts and circumstances” test); AMERCO, 96 T.C. at 38 (opinion by Judge Korner rejecting any unified “test” and remarking that the considerations “are not independent or exclusive. Instead, we read them as informing each other and, to the extent not fully consistent, confining each other’s potential excesses.”).
No set of criteria is a “test.” Lists without metes, bounds, weights, or means of resolving conflicts do not identify necessary or sufficient conditions; they never [864]*864prescribe concrete results. Perhaps a list is all we can expect when the statute is silent and both sides of a dispute have solid points. For the Commissioner is right to say that Sears does not buy insurance in the same sense as a natural person buys auto insurance, and that it transfers less risk when buying a policy from Allstate than when buying the same policy from Nationwide. Sears is right to say that Allstate sells Sears a product that passes for insurance in the industry, identical to what Allstate sells to its other clients and having economic consequences differing from a self-insurance reserve. Perhaps disputes of this kind do little more than illustrate the conundrums inherent in an effort to collect a tax from corporations, as opposed to a tax measured by the changes in wealth of corporate investors (or measured by their withdrawals for consumption, so as to encourage investment). The experts who labored during this trial to define “insurance” all would have agreed that this dispute is an artifact of the corporate income tax, which by divorcing taxation from real persons’ wealth, income, or consumption is bound to combine tricky definitional problems with odd incentives.
Suppose we ask not “What is insurance?” but “Is there adequate reason to recharacterize this transaction?”, given the norm that tax law respects both the form of the transaction and the form of the corporate structure. It follows from putting the matter this way that the decision of the Tax Court must be affirmed. For whether a transaction possesses substance independent of tax consequences is an issue of fact — something the Commissioner harps on when she prevails in the Tax Court. E.g., Yosha, 861 F.2d at 499 (citing cases). The transaction between Sears and Allstate has some substance independent of tax effects. It increases the size of Allstate’s pool and so reduces the ratio between expected and actual losses; it puts Allstate’s reserves at risk; it assigns claims administration to persons with a comparative advantage at that task. These effects are no less real than those of loans and interest payments within corporate groups — which the Commissioner usually respects even though they are occasionally recharacterized as contributions to capital. E.g., National Farmers Union Service Corp. v. United States, 400 F.2d 483 (10th Cir.1968); Crosby Valve & Gage Co. v. CIR, 380 F.2d 146 (1st Cir.1967). Hartford is a subsidiary of ITT, as Allstate is of Sears. Suppose Sears were to buy from Hartford the same policies it obtained from Allstate, and Allstate were to serve ITT’s needs. Then even the Commissioner would concede that both ITT and Sears had “insurance,” yet nothing of substance would differ — not given the Commissioner’s concession that Allstate wrote policies with standard commercial terms at competitive premiums. A trier of fact may, and did, conclude that Allstate furnished Sears with insurance.
II
PMI Mortgage Insurance Company, another part of the Sears group, writes mortgage insurance. PMI Mortgage Insurance and its own subsidiary, PMI Insurance Company (collectively PMI), insure lenders against the risk that borrowers will not pay. The Tax Court’s opinion marshals the facts, 96 T.C. at 73-85, which are unnecessary to recount at length. Two dominate: (1) The insured risk is a borrower’s default in payment. (2) Mortgage insurers insist that the lender try to collect from borrowers or realize on the collateral; until the lender has foreclosed on or otherwise obtained title to the property securing the loan (which also fixes the amount of the loss), the insurer does not pay. The last statement is a simplification. Sometimes PMI compromises with the lender in advance of foreclosure, but the policy does not require PMI to pay until the lender has good title.
Lenders must tell PMI about defaults and the steps they have taken to collect.. PMI establishes reserves for losses when one of the following occurs: (a) the property has been conveyed to the lender but not sold to a third party; (b) the property is in the process of foreclosure; or (c) the loan has been in default for four months or [865]*865more. PMI also estimates the number of loans for which one of these three things has occurred but not been reported. Such reserves for incurred but not reported (IBNR) casualties are staples of the insurance business, and the Commissioner does not contest the establishment of IBNR reserves, provided that an identical reported event would support a loss reserve deductible under § 832.
Obviously not all of these events will lead to obligations on the insurance. Borrowers may catch up on overdue payments and retire their loans. Property sold at foreclosure may generate proceeds adequate to cover the outstanding balance of the loan. Insurers, including PMI, therefore discount their reserves to reflect their experience (and the industry’s). PMI discounted too heavily, as things turned out. For 1982 PMI established year-end reserves of $35.9 million. The amounts disbursed in later years on account of these defaulted loans came to $51.5 million. So its reserve was too small. But the Commissioner believes that PMI’s reserves were too big for tax purposes. She limited the loss reserve deduction to $19.5 million for 1982, making comparable cuts for other years, and the Tax Court sustained her decisions. The court held that an “insurer cannot incur a loss until the insured has suffered the defined economic loss, to wit, after the lender takes title to the mortgaged property and submits a claim for loss.” 96 T.C. at 114. By this time, of course, there is no need for a “reserve”; payment is a current obligation. The court’s approach does not affect the taxes of insurers in a steady state but substantially increases the taxes of those with growing businesses (or growing losses) by postponing the time when the losses may be deducted.
Sears contests this' decision on two grounds: first, that § 832 does not limit loss deductions to casualties that have reached the point of being payable; second, that the Tax Court erred in deciding when an insurer’s obligation attaches. Judge Whalen (joined in this respect by Chief Judge Nims and Judge Jacobs) agreed with the former argument, concluding that the majority’s holding “is a radical departure from the annual statement method of accounting, which section 832 and its predecessors have required property and casualty insurance companies to use in reporting underwriting and investment income for Federal income tax purposes since 1921.” 96 T.C. at 114-15.
The “annual statement method of accounting” to which Judge Whalen referred is prescribed by the National Association of Insurance Commissioners, a body comprising state insurance regulators that has filed a brief as amicus curiae in support of Sears. The NAIC’s annual statement requires property and casualty insurers to take certain things into income and prescribes reserves. A mortgage insurer must include in its reserves the three categories of losses that PMI used, plus reserves for IBNR losses. The Commissioner concedes that PMI complied with the NAIC’s requirements. Federal agencies such as the Federal Housing Administration engaged in guaranteeing loans account for loss reserves exactly as PMI did. The insurance regulators believe that PMI erred, if at all, in understating its loss reserves. No surprise here. State regulators strive to assess and preserve the solvency of insurers. Accurate estimates of losses are essential to the former task, and high estimates contribute to the latter by requiring insurers to obtain additional capital or curtail the writing of new policies. Regulators therefore favor generous estimates of losses, while the federal tax collector prefers low estimates. The majority of the Tax Court stressed this when concluding that PMI could not follow the NAIC’s method: “The objectives of State regulation ... are not identical to the objectives, of Federal income taxation. State insurance regulators are concerned with the solvency of the insurer_ In contrast, Federal tax statutes are concerned with the determination of taxable income on an annual basis.” 96 T.C. at 110.
Generalities about what “[fjederal tax statutes are concerned with” do not control concrete cases. Section 832 is no ordinary [866]*866rule. It expressly links federal taxes to the NAIC’s annual statement:
(a) In the case of [a property or casualty] insurance company ... the term “taxable income” means the gross income as defined in subsection (b)(1) less the deductions allowed by subsection (c).
(b)(1) The term “gross income” means the sum of—
(A) the combined gross amount earned during the taxable year, from investment income and from underwriting income as provided in this subsection, computed on the basis of the underwriting and investment exhibit of the annual statement approved by the National Association of Insurance Commissioners. ...
(b)(3) The term “underwriting income” means the premiums earned on insurance contracts during the taxable year less losses incurred and expenses incurred.
(b)(5)(A) The term “losses incurred” means losses incurred during the taxable year on insurance contracts computed as follows:
(i) To losses paid during the taxable year, deduct salvage and reinsurance recovered during the taxable year.
(ii) To the result so obtained, add all unpaid losses ... outstanding at the end of the taxable year and deduct all unpaid losses ... outstanding at the end of the preceding taxable year.
This quotation includes changes made in 1988, but these do not affect the current dispute. Section 832(b)(1)(A) requires an insurer to use “the underwriting and investment exhibit of the annual statement approved by the National Association of Insurance Commissioners” to determine its “gross income.” Contrary to usual notions of “gross income,” this concept in § 832 does not denote all inflows of - revenue. Instead it refers to “premiums earned” (a premium is not “earned” until the period for which it purchases coverage occurs) less “losses incurred.” For purposes of § 832, then, “gross income” is a version of net earned income. Both the “premiums earned” and “losses incurred” go into determining “gross income” — which is to be “computed on the basis of the underwriting and investment exhibit of the annual statement approved by the National Association of Insurance Commissioners”. State insurance commissioners’ preferences about reserves thus are not some intrusion on federal tax policy; using their annual statement is federal tax law. See Brown v. Helvering, 291 U.S. 193, 201, 54 S.Ct. 356, 360, 78 L.Ed. 725 (1934): “[T]he deductions allowed for additions to the reserves of insurance companies are technical in character and are specifically provided for in the Revenue Acts. These technical reserves are required to be made by the insurance laws of the several States.”
True enough, the definition of loss reserves in § 832(b)(5) does not refer to the annual statement. Yet subsection (b)(5) losses are a component of subsection (b)(1) income, which is to be computed according to the NAIC’s statement. It is scarcely possible to use the statement when determining one but not the other. Although it is not impossible — almost nothing is impossible in tax law — divorcing (b)(5) losses from the annual statement computations would make no sense in terms of the structure of the statute or its genesis. Subsection (b)(5) prescribes a method of toting up losses derived almost verbatim from the annual statement used in 1921, when Congress enacted the provision.
If annual statements were to depart from an effort to approximate actual “losses” then subsections (b)(1) and (b)(5) might come into conflict. This occurred when states required insurers to mark up their loss reserves by a percentage. The Commissioner objected to the deduction of these marked up losses, issuing regulations in 1943 and 1944 requiring insurers to use experience, and not formulas prescribed by state rules, as the basis of loss reserves. Modified versions of these regulations are still in force but no longer present the insurers with conflicting state and federal demands. In 1950 the NAIC came ’round [867]*867to the Commissioner’s point of view, changing its annual statement so that both federal and state governments require insurers to reserve “only actual unpaid losses ... stated in amounts which, based upon the facts in each case and the company’s experience with similar cases, represent a fair and reasonable estimate of the amount the company will be required to pay.” Treas. Reg. 1.832-4(b). Charles W. Tye, The Convention Form and Insurance Company Tax Problems, 6 Tax L.Rev. 245 (1951), narrates the history of this dispute and the details of its resolution. PMI used actual cases to generate its loss reserves, and in the event underestimated losses; it complied with both the NAIC’s requirements and the Treasury’s regulations. Having followed the NAIC’s annual statement approach, PMI is entitled to deduct the loss reserves so computed.
For what it is worth, we believe that PMI would be entitled to prevail under the regulation independent of the requirements of the NAIC’s annual statement. The regulation says “actual unpaid losses” but omits any requirement that these losses be quantified and immediately payable. Once an obligation is quantified, an accrual-basis taxpayer may deduct it. Yet § 832 and the regulation suppose that insurers may deduct losses denied to any old accrual-basis taxpayer, a supposition the Supreme Court confirmed in General Dynamics when holding that an employer paying for its employees’ medical care without an insurer’s intermediation could not deduct IBNR expenses (in this context, the cost of medical services already rendered to employees but for which the employer did not have bills in hand).
Consider some standard issues in establishing reserves. A policy of auto insurance requires the issuer to pay if its insured is at fault in a collision. An accident occurs during December. May the insurer add to its reserves? The liability is not fixed, for the insurer is not legally obliged to pay until a court determines that its policy-holder was at fault (or the underwriter so concedes), and even then the firm may not be called on to pay if the loss turns out to be less than the deductible or the victim collects from his own carrier, which decides not to pursue the other driver’s carrier. It may take years before the amount of the loss is quantified and the negligent driver is identified. Yet reserves established for such a case meet the regulatory definition of actual, case-based losses, and it would be insane of an insurer not to establish reserves for such casualties. Or consider health insurance. An insured has a heart attack on December 31. Medical care will be required over the next months (or years), and the insurance policy conditions the obligation to pay on receipt of a physician’s bill at rates usual and customary in the vicinity (with a provision for arbitration if the fee seems high or the medical services unnecessary). Once again it may be some time before services have been rendered and billed at rates agreeable to the carrier. Does it follow that the insurer must wait till it receives the bill before establishing a reserve? At oral argument counsel for the Commissioner answered “yes,” but the Commissioner cannot mean it, for this answer collapses all distinction between “reserves” and bills payable by return mail. Cf. Harco Holdings, Inc. v. United States, 969 F.2d 440, 442 (7th Cir.1992).
Just so with the Tax Court’s conclusion about mortgage insurance. It has confused quantification of the loss, which does not occur until the lender tenders title, with the occurrence of the covered loss. Perhaps it seems artificial to speak of a borrower’s failure to make a few payments as a “loss.” The borrower may catch up, or the sale of the property may reimburse the lender. Default is not an immediate casualty in the sense that a collision between two automobiles crushes the cars (and people) on the spot, and it does not portend outlays with the high probability that a myocardial infarction does. Yet the acid test is whether the default leaves the insurer responsible for payment. Let us suppose that PMI issues a policy for 1982 only, and the borrower omits the last four payments of the year. The lender neglects to renew the policy (or purchase a substitute) for 1983. Eventually the lender forecloses [868]*868and sends PMI a bill. Must PMI pay? The answer is yes; the default is the event triggering coverage under the policy. (Neither the Commissioner nor the Tax Court disagrees with PMI’s representations about its obligations under the policy.) Thus to state its statutory “gross income” for 1982 accurately, PMI must take into income the premiums earned during 1982 and exclude a reserve for losses attributable to those premiums, including the bills that will straggle in during future years on account of defaults that began in 1982. The Tax Court’s observation that federal law calls for “determination of taxable income on an annual basis”, 96 T.C. at 110, turns out to support PMI, once we see that default is the event triggering coverage under the policy and leaving the insurer on the hook, waiting to see how things turn out, even if it never receives another penny in premiums.
Corporate taxation teems with artificial and formal distinctions, and the taxation of insurers has more than its share of them. Whether § 832 is attributable to some finely honed sense of the economics of the insurance business or to political pressure is not for us to say. Provisions of the Internal Revenue Code do not conflict with “tax policy,” as the Commissioner seems to believe. They are tax policy and are to be enforced. Usually this enlarges the revenue. E.g., Holywell Corp. v. Smith, — U.S. -, 112 S.Ct. 1021, 117 L.Ed.2d 196 (1992); INDOPCO v. CIR, — U.S. -, 112 S.Ct. 1039, 117 L.Ed.2d 226 (1992). An Internal Revenue Service eager to dish out the medicine of literalism-must be prepared to swallow it. Sears is entitled to prevail on both branches of this case. The judgment of the Tax Court is affirmed with respect to the Allstate dispute and reversed with respect to the PMI dispute. The case is remanded for the redetermination of the deficiency in accord with this opinion. The Tax Court is free to consider the Commissioner’s argument, which it did not need to reach before, that PMI’s returns for 1980 and 1981 did not use a proper case-based method of approximating its loss reserves.