Diane Cutting was seriously injured in an automobile accident. She has incurred $90,000 in medical expenses alone, and expects to incur additional medical expenses in the future. An uninsured-motorist policy paid her $126,000 and she recovered another half million dollars on a products-liability claim, but as she reckons her total damages from the accident at $1 million there is a considerable shortfall. Her husband's employer, Jerome Foods, Inc., has an unfunded, company-administered employee benefits plan that covers spouses. The Cuttings submitted Mrs. Cutting's medical expenses to the plan for payment. The plan was willing to pay--provided the Cuttings agreed to reimburse it for any amounts that they had received from other sources. The basis of this demand was the subrogation clause of the benefits plan, which provides that by accepting any payment of plan benefits the covered employee or dependent "agrees that the Plan shall be subrogated to all claims, demands, actions and rights of recovery of the individual against any third party or any insurer, including Workers' Compensation, to the extent of any and all payments made or to be made hereunder by the Plan." The Cuttings refused to reimburse the plan, contending that a right of subrogation does not arise unless the covered individual has been made whole, in which case the payment of benefits under the plan would confer a windfall. If it is assumed that Diane Cutting really did sustain a loss of $1 million of which more than $900,000 is for items of loss or expense other than medical expenses, then since she has recovered less than $650,000 the effect of subrogation would be to reduce her recovery of nonmedical expenses by the exact amount of the plan payments, so that she would derive no benefit in fact from the plan. Although there would be a sense in which her medical expenses had been fully paid (by the product manufacturer and the uninsured-motorist insurer), it would be a misleading sense. The "payment" would be out of money that would otherwise have been hers to apply to other items of loss caused by the accident that gave rise to the medical expenses.

The company rejected the Cuttings' interpretation of the subrogation clause, precipitating this declaratory judgment action by the Cuttings. Having removed the suit to federal court because the plan is governed by the Employee Retirement Income and Security Act (ERISA), 29 U.S.C. §§ 1001 et seq., Jerome Foods moved successfully for summary judgment. 820 F. Supp. 1146 (W.D.Wis.1992).

The plan provides that "all decisions concerning the interpretation or application of this Plan shall be vested in the sole discretion of the Plan Administrator," that is, Jerome Foods. Read literally, this terminology would extinguish all judicial review of refusals by Jerome Foods to pay benefits. The company concedes, however, that it should not be read literally--that it is not a license to make arbitrary or capricious decisions on applications for benefits. Cf. Lister v. Stark, 942 F.2d 1183, 1188 (7th Cir.1991). Were it otherwise, the coverage provided by the plan would be, not illusory exactly but noncontractual, because the company could turn down an application for benefits on whim. It might sometimes be tempted to do so. The plan is unfunded, with the result that every penny paid out in plan benefits comes out of the company's coffers; and even with funded or insured welfare plans, the employer has a financial stake in limiting the payment of benefits.

Of course, a company that is capricious in its bestowal of benefits may have to pay higher wages to compensate its workers for the anticipated reduction in the certainty and value of their benefits, or it may incur the potentially costly ill will of the workers. Some workers might prefer to rely on their employer's long-term self-interest in playing fair with benefits than to insist on a costly regime of legal rights for which they pay indirectly in lower wages or more limited benefits, just as many workers appear to prefer jobs in which they are employees at will to tenured positions as civil servants. We are not certain that there is any legal impediment to a plan's forbidding judicial review of the plan administrator's decisions. ERISA provides that "a civil action may be brought by a participant or beneficiary ... to recover benefits due," 29 U.S.C. § 1132(a)(1)(B), but if the plan entrusts the determination of what is due to the unreviewable discretion of the plan administrator, it can be argued that no benefits will be due unless the administrator awards them. And although the Supreme Court held in Firestone Tire & Rubber Co. v. Bruch, 489 U.S. 101, 115, 109 S. Ct. 948, 956, 103 L. Ed. 2d 80 (1989), that the default standard of judicial review of a plan administrator's decision is de novo review, it also held that if the plan gives the administrator discretion to construe its provisions the court will defer. Id. at 111, 115, 109 S. Ct. at 954, 956; Lister v. Stark, supra, 942 F.2d at 1187-88. Not defer completely, however--defer in the sense conveyed by such familiar and possibly synonymous terms as "abuse of discretion" and "arbitrary and capricious" (between which the Court found it unnecessary to choose), for which clear error may be a further synonym. Id. at 1188; Haugh v. Jones & Laughlin Steel Corp., 949 F.2d 914, 916-17 (7th Cir.1991). It is one thing to interpret a provision that vests discretion in a plan administrator as commanding judicial deference within reason, another thing to interpret it as commanding complete deference. The indications that the latter was intended would have to be pretty conclusive to persuade us, and we agree with the parties that the reference to "sole discretion" is not (quite) conclusive evidence of such an intention and therefore should not be interpreted to strip the plan's beneficiaries of all legal remedies, especially given the conflict of interest inherent in an unfunded employer-administered plan. (Compare the hostility of trust law to efforts by a trustee to place himself beyond liability for profiting from a breach of trust. Restatement (Second) of Trusts § 222(2) (1959).) Otherwise the employees would have no contractual protection.

It is true, as we have pointed out, that employees often bargain away contractual protection. That is of the essence of the regime of employment at will, which is still the dominant regime for employment in this country. But this is only to say that employees normally will not pay for job insurance beyond what arrangements for unemployment compensation and severance pay provide them. They will pay for medical insurance and they usually want more assurance of insurance than an unenforceable promise from their employer would give them. Although an employer's concern for employee relations and good will will inhibit him from indulging in wholly irrational self-serving interpretations, it is not a complete guarantee. A rational employer will trade off the cost of such an interpretation against the gain in reduced expense, which might be critical, especially for employers in a financially parlous state. A good business reputation is not a very valuable asset to a firm that has little confidence of being able to remain in business long enough to profit from that reputation. Long-run self-interest is a weak constraint on firms that have poor long-run prospects.

So the Cuttings are entitled to judicial review of the company's ruling. But they are wrong to argue that the standard of judicial review is de novo--that we should give no weight to the company's interpretation. That as we said is the default standard. If the plan itself vests discretion in the administrator--if as here it gives the administrator a long leash--the courts pull back and defer broadly although not totally to the administration's determination, upending it only if persuaded that the administrator acted unreasonably. Firestone Tire & Rubber Co. v. Bruch, supra, 489 U.S. at 115, 109 S. Ct. at 956.

The interpretive question is sufficiently close in this case that a deferential standard of review dictates affirmance of the plan's decision, and hence of the district court's decision. We may put to one side Jerome Foods' argument that all that the Cuttings are contending is that we should apply Wisconsin subrogation law to this case. Wisconsin is indeed a state that recognizes the "make whole" qualification to a subrogee's rights; but the Cuttings concede as they must that ERISA preempts state law dealing with the interpretation of an ERISA plan, unless, as is not the case here, the plan involves the purchase of an insurance policy as the method of providing plan benefits. See FMC Corp. v. Holliday, 498 U.S. 52, 111 S. Ct. 403, 112 L. Ed. 2d 356 (1990), holding a state subrogation rule preempted; Provident Life & Accident Ins. Co. v. Linthicum, 930 F.2d 14 (8th Cir.1991) (per curiam) (ditto). The Cuttings do not ask us to apply state law. They ask us to adopt a federal common law rule to the effect that rights of subrogation are enforceable only after the plan beneficiary has been made whole for the loss giving rise to the claim for benefits. That the federal rule might coincide with the law of Wisconsin is of no moment--especially since the Cuttings argue that the rule for which they contend is the universal rule applied in subrogation cases, in which event it is necessarily identical to the Wisconsin rule.

There is no doubt about the authority of the federal courts to create common law for use in ERISA cases. Pilot Life Ins. Co. v. Dedeaux, 481 U.S. 41, 56, 107 S. Ct. 1549, 1557, 95 L. Ed. 2d 39 (1987); Fox Valley & Vicinity Construction Workers Pension Fund v. Brown, 897 F.2d 275, 281 (7th Cir.1990) (en banc). We suppose that even Jerome Foods would concede that these courts have to adopt some interpretive principles, even if only implicit ones, in order to construe ERISA plans, since ERISA sets forth no principles of interpretation of its own. And those principles that the judges devise or adopt to guide their interpretations are therefore common law, that is, judge made, principles.

As it happens, the so-called universal rule of subrogation for which the Cuttings contend is a rule of interpretation. No one doubts that the beneficiary of an insurance policy or (as here) an employee welfare or benefits plan can if he wants sign away his make-whole right. Garrity v. Rural Mutual Ins. Co., 77 Wis. 2d 537, 253 N.W.2d 512, 515 (1977); Lyon v. Hartford Accident & Indemnity Co., 25 Utah 2d 311, 480 P.2d 739, 744 (1971), overruled on other grounds in Beck v. Farmers Ins. Exchange, 701 P.2d 795, 798 (Utah 1985); Wimberly v. American Casualty Co., 584 S.W.2d 200, 204 (Tenn.1979). The right exists only when the parties are silent. It is a gap filler. We need not consider the judges' power to impose substantive or remedial, as distinct from merely interpretive, common law principles, on ERISA plans; need not consider, that is, the extent to which common law reasoning can be used to enlarge (or contract) the rights of an ERISA beneficiary or participant. The scope of that power is uncertain, Martin v. Consultants & Administrators, Inc., 966 F.2d 1078 (7th Cir.1992); Greany v. Western Farm Bureau Life Ins. Co., 973 F.2d 812, 821-22 (9th Cir.1992), because of the possibility of conflict between a specific substantive or remedial principle, such as equitable estoppel, and the substantive or remedial principles set forth in ERISA itself, such as the requirement (with which equitable estoppel might be thought inconsistent) that all plan benefits be in writing. Cf. Decatur Memorial Hospital v. Connecticut General Life Ins. Co., 990 F.2d 925, 926-927 (7th Cir.1993). We are dealing only with an interpretive principle.

It is difficult to say whether the "make whole" interpretive principle is predominant, let alone universal. The treatises do not agree on the point. Compare 16 Mark S. Rhodes, Couch on Insurance Law § 61:64 at pp. 145-47 (2d rev. ed. 1983), and Allan D. Windt, Insurance Claims and Disputes: Representation of Insurance Companies and Insureds § 10.06 at pp. 533-39 (2d ed. 1988), with Robert E. Keeton & Alan I. Widiss, Insurance Law: A Guide to Fundamental Principles, Legal Doctrines, and Commercial Practices § 3.10(b) at pp. 233-40 (1988). Although all but one modern case that we have found give lip service to it as the default rule (the exception is In re Estate of Scott, 208 Ill.App.3d 846, 153 Ill. Dec. 647, 649, 567 N.E.2d 605, 607 (1991)), many allow it to be overridden by a boilerplate subrogation clause. (With the Garrity, Lyon, and Wimberly cases, cited earlier, compare Foremost Life Ins. Co. v. Waters, 415 Mich. 303, 329 N.W.2d 688 (1982) (per curiam), and Peterson v. Ohio Farmers Ins. Co., 175 Ohio St. 34, 191 N.E.2d 157, 159 (1963).) The rule hardly exists, as a practical matter, in the states that allow this.

It is difficult to say which is the better rule. In the usual case the insured incurs a medical expense for which he is promptly reimbursed by the insurance company (or equivalent), which then exercises its right of subrogation and goes off on what sometimes proves to be a wild goose chase after a possible tortfeasor to whom the expense can be shifted. The insured is out of the picture--well, not entirely, because he is entitled to any excess recovery by the insurance company. Subject to this qualification, of which more later, the risk of an uncertain recovery is shifted from the individual to a specialist in bearing risk and enforcing claims--the insurance company. The price of the insurance is thereby reduced (or, equivalently, the insured can buy more insurance for the same price) because the insurance company acquires a contingent claim by virtue of the subrogation clause. Without subrogation, a part of the risk is shifted back to the insured. He pays more for the insurance because he retains, rather than (in effect) sells, a right to obtain through litigation a recovery that may actually exceed the actual loss that (after receiving the insurance proceeds) he suffered. This is because, by virtue of the collateral benefits rule, a tort plaintiff need not deduct from the judgment he receives in his tort suit compensation for the same loss which he may have obtained from another source, such as an insurance policy.

A legal right, for example a tort victim's right to sue his tortfeasor for damages, is like a lottery ticket. Its payoff is highly uncertain. Subrogation, consistent with the objective of insurance, transfers the lottery ticket from the individual policyholder to the collectivity of policyholders. The risk is spread, and by being spread eliminated or at least greatly reduced. Subrogation also eliminates the incentive to careless behavior created by the collateral benefits rule, which by enabling the insured victim of a tort to recover for the same loss from both the insurance company and the tortfeasor may (depending on the adequacy of tort damages) make the tort victim better off than he would have been had there been no accident.

So it is not surprising that subrogation clauses are a common feature of insurance policies. But the "make whole" issue is not about whether insurance subrogation is in general a good thing, as it doubtless is. A "make whole" rule does not enable the insured to gamble on double recovery, as he would be doing without subrogation. The rule comes into play only if without it the operation of the subrogation clause would, by making the insured in effect pay back the policy benefits with money he has received for another loss caused by the injury that gave rise to the policy claim, leave him with an uncompensated injury. That would be the consequence of subrogating Jerome Foods to Mrs. Cutting's tort claim to the extent of the plan's medical benefits; the uncompensated nonmedical costs that she incurred as a result of the accident would be $90,000 greater if subrogation is permitted. One can see why many state courts have thought that the inclusion of a boilerplate subrogation clause was not intended to have this effect--the effect, practically speaking, of curtailing coverage. To put this differently, rejection of "make whole" makes subrogation a lot like assignment. If insurance contracts required the insured to assign any tort claim he might have to the insurer, the price of the insurance would be lower but effective coverage would also be lower. The insured would recover only the policy limits, and not his full damages, even in a case in which a judgment had been secured against the tortfeasor, and collected, for those damages.

These arguments need not be thought decisive in favor of the make-whole rule. It is true that rejecting make whole would bring subrogation closer to assignment, but that would not necessarily be a bad thing. Assignment, by shifting the insured's tort rights to the insurance company, reduces the price of insurance and thus enables the insured to obtain more coverage, in effect trading an uncertain bundle of tort rights for a larger certain right, which is just the sort of trade that people seek through insurance. (The counterargument is that the make-whole rule is most likely to apply in a case of catastrophic loss, such as the present, where full insurance is rare.) It can also be argued against the make-whole rule that it is administratively more complex, requiring the medical insurer to calculate the insured's total medical and nonmedical loss, and therefore that it makes insurance (or its equivalent in this case) more expensive to the insured--and makes it more expensive to him for the additional reason that he will have to pay for the additional coverage that the rule in effect provides. And no one's demand for insurance is unlimited. The counterargument on the administrative-complexity front is that subrogation (which the make-whole rule would eliminate in cases such as the present) involves an agency problem because of the insured's entitlement to any excess recovery that the insurer-subrogee obtains. The insurer, having no (or very little) interest in making money for the insured, may fail to press hard for an excess recovery. See Keeton & Widiss, supra, § 3.10(c)(2) at pp. 242-44. To which it can be replied that the make-whole rule simply shifts the disincentive to press hard from the insurer to the insured.

Fortunately we need not attempt to decide in this case whether the merits of the rule are sufficient on balance to warrant its use as a principle of interpretation of ERISA plans. Because, as the Cuttings concede, the make-whole rule is just a principle of interpretation, it can be overridden by clear language in the plan. Jerome Foods contends that the language is clear. The plan document does state rather flatly that the plan shall be subrogated to "all claims" by the covered individual against a third party to the extent of "any and all payments" made (or to be made) by the plan. But the language is no stronger than that in cases which have held that more was required to override the make-whole principle. And one of those cases (Garrity ) is a Wisconsin case and Jerome Foods is a Wisconsin corporation (it both is incorporated in Wisconsin and has its principal place of business in that state), and this may be a clue to meaning, even though Wisconsin law does not govern the case. But at this point in the forensic badminton game the standard of judicial review becomes critical and decides the case in favor of Jerome Foods. For we cannot say that the company was unreasonable in interpreting this plan as disclaiming the make-whole principle. Had the Cuttings not refused to sign the reimbursement agreement, in effect the plan would have been advancing them their medical expenses against the possibility of a tort recovery sufficient to cover those expenses, and indeed such a recovery was obtained. This is as we have said a more limited form of insurance than is created by a make-whole provision, but Jerome Foods may not have been willing to provide--or the Cuttings to pay for--more generous coverage. We shall never know. We cannot say, however, that the administrator was unreasonable in concluding that the plan had disclaimed any obligation to make the Cuttings whole.

AFFIRMED.