Until it became a subsidiary of Sundstrand in 1984, Sullair Corporation had a floor- offset pension plan. Distinctive features of that plan still apply to persons who worked at Sullair before the acquisition. A floor-offset plan uses a defined- benefit structure (with pension payments linked to years of work and high salary) to buffer the uncertainty of a defined- contribution system (where pension payments depend on the performance of investments in each employee's account). See Brengettsy v. LTV Steel (Republic) Hourly Pension Plan, 241 F.3d 609 (7th Cir. 2001); Regina T. Jefferson, Rethinking the Risk of Defined Contribution Plans, 4 Fla. Tax Rev. 607, 668-71 (2000). Sullair's pre-acquisition defined-contribution plan was an Employee Stock Ownership Plan. Employees were entitled to buy Sullair stock, which the esop would hold for their account. An esop is a high-risk investment: employees' retirement wealth is undiversified, so if something happens to the firm a retiree can end up with little beyond Social Security benefits, while if the firm prospers he may live in the lap of luxury. A defined-benefit component ensures that retirees have some supplement to Social Security even if the employer (or the stock market) goes south. The principal question in this class action is how the plans interact when an employee quits Sullair after pension benefits have vested, but before retirement age.

Sullair's floor-offset plan gives retirees the greater of the defined- benefit amount and the value of an annuity that could be purchased with the stock held for the retiree's benefit by the esop. If the annuity that could be purchased with the esop stock is less than the defined-benefit amount, then the defined-benefit plan pays the difference, topping up the employee's pension. Two aspects of this arrangement give rise to the question at hand. First, the employee does not surrender the esop units in order to receive a supplemental payment from the defined-benefit plan; the employee is free to cash out the esop and buy an annuity but also is free to keep the stock and continue to bear the risk. Second, and essential in light of the first, the coordination between plans is done with a hypothetical annuity, which presents the question how the monthly annuity benefit will be calculated.

An example shows how the floor-offset arrangement works. Suppose that a given employee who retires at age 65 would be entitled to a $2,000 monthly pension from the defined-benefit plan. This serves as the floor. If this employee has a $100,000 balance in the esop, the plan administrator determines how large an annuity the $100,000 could purchase. That would be about $900 per month (assuming 7% annual interest and a 15-year life expectancy). Sundstrand then would offset the $900 hypothetical annuity and pay the retiree $1,100 per month; the esop balance is the retiree's to annuitize, hold, or invest as he pleases. If, however, the employee has a $300,000 esop balance on retirement, an amount that could purchase an annuity of about $2,700 monthly, then the payment from the defined-benefit plan would be zero. Again the retiree retains the ability to draw down or invest the esop balance. Everything works smoothly when the employee starts receiving pension benefits immediately on leaving Sullair.

What happens when there is a gap between leaving Sullair and retiring? The plan's administrative committee calculates the monthly defined-benefit amount based on the employee's length of service and terminal salary. That much is uncontroversial. What about the defined- contribution offset? The plan takes the value of the employee's esop account on the day he left Sullair's employ and places that amount in a hypothetical savings account, at the same interest rate used in the annuity calculations, until the day the employee becomes eligible for retirement; at that point the balance is annuitized, and the resulting monthly benefit subtracted from the defined-benefit amount. So if an employee quits at age 45, with $100,000 in the esop account, and plans to retire at age 65, the plan assumes that this balance will compound for 20 years at the going rate (7% in our example), producing a kitty of almost $390,000 by retirement. This sum, when annuitized, is about $3,500 per month, so the monthly defined- benefit is zero (but, as always, the employee can do what he pleases with the esop balance from age 45 onward). This process is equivalent to asking what monthly annuity an employee could buy with $100,000 today, when payments on the annuity would be deferred for 20 years. The parties call this approach the "deferred rate" calculation. The other approach, which the plaintiffs favor and call the "immediate rate" calculation, asks how large an annuity a 45-year-old could buy with $100,000 for immediate commencement. That would be about $640 monthly--less than the $900 calculated in our first example because payments would be expected to last 35 years instead of 15. Under the immediate-rate calculation, then, the retiree's benefit from the pension plan would be $1,360 monthly, starting at age 65. Meanwhile the esop balance is likely to grow to $390,000 by age 65, so this person's total monthly benefit will be about $4,900 (the $1,360 from the plan, plus the $3,500 monthly that could be obtained by using the esop balance at age 65 to purchase an annuity). The difference between deferred and immediate calculation can be substantial (though there will be no difference if even the immediate-rate calculation exceeds the monthly pension generated by the defined-benefit component). The functional question is: Who gets the benefit of interest between quitting and retirement age, the employee or the plan?

Any pension plan giving retirees the greater of two amounts, as opposed to the sum of these amounts, can be described as confiscating the difference. That's the nature of a floor-offset plan: the retiree always "loses" the defined- contribution balance (principal and interest) up to the floor in the defined- benefit plan. What the employee loses, the plan receives. Using the esop to fund part of the defined-benefit promise makes any given level of promised benefits cheaper to the employer and so may increase the fallback pension. But under erisa that is neither here nor there. The Employee Retirement Income Security Act does not require employers to establish plans that are particularly favorable to employees. There is no fiduciary duty to employees when establishing plans' provisions. See Hughes Aircraft Co. v. Jacobson, 525 U.S. 432 (1999); Lockheed Corp. v. Spink, 517 U.S. 882 (1996). The employer's fiduciary duty, as plan administrator, is to implement faithfully the provisions of the plan as written. Thus we arrive at the critical language of the Sullair plan (we quote the 1981 version, which has not been altered substantively):

The amount of monthly pension which could be provided on a straight life annuity basis by application of an amount equal to the fair market value as of such specified date of the Participant's vested interest in the balance credited to his account under the Sullair Corporation Employee Stock Ownership Plan, and, for this purpose, such amount of monthly pension shall be determined on the basis of the annuity purchase rates in effect as of such specified date under and as set forth in the Group Contract.

So the hypothetical calculation uses the "annuity purchase rates"--whatever they are. In 1981 and for many subsequent years they were set by Bankers Life (now Principal Mutual Life Insurance Co.), which underwrote the plan with a group annuity contract. The first time the issue came up, Bankers Life used an immediate-rate approach, but it reversed itself before any benefits were paid and told the Plan Benefit Committee in May 1985 that a deferred-rate approach should be used. The Committee agreed, and a deferred-rate approach has been in force ever since.

Plaintiffs observe that immediate-rate calculations are used for persons still on Sullair's payroll at retirement, and they demand the same for themselves. Moreover, when the plan uses a deferred- rate calculation for persons who leave Sullair before retirement age, the floor- offset system loses some or all of its risk-buffering function. The deferred- rate calculation assumes that the balance in the esop will grow at some steady rate to retirement age. If it grows less (or even shrinks), the employee bears the whole loss, without any assistance from the defined-benefit component, although an employee who works at Sullair through retirement receives the assurance that if the esop's value collapses at the last moment, the defined-benefit floor still is there. (A person who leaves Sullair before retirement does have the option, however, of rolling the esop balance over into a more diversified fund, thus protecting against firm-specific risks.)

If immediate-rate calculations are used, however, then employees who leave Sullair before retirement age obtain a big advantage over those who stay--for the floor-offset plan assuredly deprives lifetime employees of some or all value of growth in the esop account. Think again about our hypothetical employee, who accumulates a value of $100,000 in the esop by age 45. If the employee quits and retires 20 years later, an immediate-rate calculation provides that person with the full value of the esop plus $1,360 per month at age 65. Now suppose that the person decides to stay at Sullair through retirement and does not make further investments in the esop. If the value of the esop account grows at the assumed 7% rate, it will be worth $390,000 at age 65, and the monthly kicker from the defined-benefit component of the plan will be zero. Exactly the same investment strategy--accumulate $100,000 by age 45, invest nothing for the next 20 years--has substantially different effects, under plaintiffs' preferred approach, depending on whether the person leaves Sullair or stays to retirement. What sense does that make? Perhaps more to the point, unless it is trying to encourage early retirement (which it isn't) why would Sullair want a pension plan that rewards employees for quitting? Cf. McNab v. General Motors Corp., 162 F.3d 959 (7th Cir. 1998). Employers can structure their plans to make it worthwhile for employees to leave, but firms usually want to encourage employees who have developed valuable skills and knowledge to remain-- and, not incidentally, encourage them to develop such firm-specific skills in the first place. Back-loading of compensation (through salary that generally increases with age as well as through defined- benefit pension plans) is a principal way to do this. See Edward P. Lazear & Robert Moore, Pensions and Turnover, in Pensions in the U.S. Economy 163-87 (Zvi Brodie, et al., eds. 1988); Edward P. Lazear & Sherwin Rosen, Pension Inequality, in Issues in Pension Economics 341 (Zvi Brodie, et al., eds. 1987). Plaintiffs' approach, by contrast, would produce a plan that treated workers staying through retirement age as suckers.

The need to use a different calculation strategy (immediate-rate for those who retire immediately, deferred-rate for those who retire later) in order to achieve an economically identical outcome offers very strong support for the plan's decision. Any remaining doubt is resolved by the fact that the plan contains a strong grant of discretion--in both interpretation and application. Section 8.01 of the 1981 plan provides:

The Plan Administrator will have complete control of the administration of the Plan, subject to the provisions hereof, with all powers necessary to enable it properly to carry out its duties in that respect. Not in limitation, but in amplification of the foregoing, it will have the power to construe the Plan and to determine all questions that may arise hereunder, including all questions relating to the eligibility of Employees to participate in the Plan and the amount of benefit to which any Participant, Beneficiary, spouse or Contingent Annuitant may become entitled hereunder. Its decisions upon all matters within the scope of its authority will be final.

This meets the requirements of Herzberger v. Standard Insurance Co., 205 F.3d 327, 331 (7th Cir. 2000), for discretion in the plan's interpretation. It was on the basis of the administrator's discretion that the district judge granted summary judgment on defendants' favor. 2000 U.S. Dist. Lexis 7273 (N.D. Ill. May 23, 2000). We agree with the district court's conclusion that the Committee's decision was not arbitrary or capricious (the right standard when the administrator has interpretive discretion)--and we would have reached the same construction as an independent matter if review were plenary.

Plaintiffs protest that deference is due to one of their number, rather than to the plan's administrator. According to plaintiffs, William R. White, the lead plaintiff, drafted the plan. Surely he is best suited to interpret it, plaintiffs insist. Not so. Self-interest would be disqualifying (White can't demand a right to calculate his own benefits), but we do not get that far. Section 8.01 gives interpretive discretion to the Plan Administrator, not to William R. White. The power goes with the office. A Secretary of Labor who drafts and promulgates a rule will have leeway to interpret its language only as long as she remains in office; when a new Secretary is confirmed, interpretive discretion comes with the job. The ex- Secretary may write op-ed pieces trying to influence the rule's application, but the elbow room that goes with delegated authority belongs to the incumbent, not to the author. Senator Wagner may have written the National Labor Relations Act, but the current members of the National Labor Relations Board (and sitting federal judges) hold the power of interpretation. For the same reason, it matters not whether the interpretation changed in 1985, after Sundstrand acquired Sullair. Whoever sits on the Plan Committee today may change matters yet again; the discretion created by sec.8.01 was not abolished in 1984 but passed to new holders.

Plaintiffs raised several arguments in addition to the immediate-rate claim, but several of these washed out before the suit began when the plan's administrative committee took a fresh look at its calculations, agreed with the plaintiffs' position, and raised their retirement benefits (and those of other Sullair employees) accordingly. Other claims survived and became the subject of this suit, though it is hard to see why. For example, plaintiffs contend that the value of phantom stock benefits (bonuses calculated by changes in the value of Sullair's stock) count as compensation for the purpose of calculating the defined-benefit component. Defendants responded that this may or may not be so, but that because none of the plaintiffs received phantom-stock compensation there was no reason to adjust their pensions. Plaintiffs ignore that problem on appeal. Plaintiffs did receive some profit- sharing payments and contend that these should be included in the base used to calculate the defined-benefit amount, but the plan documents explicitly exclude this possibility, as the district court correctly held. And the court was entirely right to deny plaintiffs' post- judgment motion, under Fed. R. Civ. P. 60(b)(3), accusing the defendants of fraud during discovery. Of this there is no evidence--and as this opinion demonstrates the outcome of the case depends on the plan's terms and the administrator's discretion, not on who said (or wrote) what to whom many years ago.

After entering its judgment on the merits, the district court ordered plaintiffs to pay about $18,000 in costs under Fed. R. Civ. P. 54(d)(1). 2000 U.S. Dist. Lexis 15335 (N.D. Ill. Oct. 19, 2000). The district court made the eight representatives jointly and severally liable for this amount. Plaintiffs do not contest the amount (a substantial reduction from the bill of costs defendants submitted) but say that liability should be individual rather than joint--and that the right denominator is the number of members in the class, rather than the number of representative plaintiffs. This class has a few more than 160 members, which implies a costs award of approximately $110 against each of the representatives. Any award against the absent class members would not be collectable--not only because the defendants have said that they would not try to do so, but also because execution would not be lawful. Class members were not given notice and an opportunity to opt out of the case, and it is impossible to see how conscripts who did not even know of the case's existence, let alone have an opportunity to distance themselves from it, could be required to pony up. Neither the federal rules nor the due process clause of the fifth amendment would tolerate such a judicial order. So the upshot of plaintiffs' rule would be that the defendants must bear most of their own costs: the larger the class, the greater the proportion of costs that prevailing defendants must bear.

Yet Rule 54 says that the prevailing party recovers costs, and nothing in Rule 23 suggests that cost-shifting is inapplicable to class actions. What justification could there be for leaving prevailing defendants out of pocket? Our eight representative plaintiffs observe that they do not want to bear the high expenses of litigation, and that protestation is easy to believe--but Sundstrand does not want to bear it either, and there is no justification for shifting the costs to the pension plan and forcing fellow retirees to bear the costs of the eight plaintiffs' mistaken litigation choices. Eight persons caused this litigation to be brought, caused the costs to be incurred, and should make the prevailing party whole. Now it's true, as the plaintiffs stress--and as we recognized in Rand v. Monsanto Co., 926 F.2d 596 (7th Cir. 1991)--that class actions are designed to aggregate claims of many persons with small stakes, and that a representative who has himself only (say) $1,000 to gain from success will be unwilling to carry the load for the rest of the class. Even if the odds of winning are favorable, few people would start a case where their maximum gain is $1,000 yet they could be hit with a bill for $20,000 or $50,000 in costs if the defendants prevail. Demanding that representative plaintiffs personally cover all costs could be the demise of consumer class actions. The stakes in this case are considerably larger than $1,000 per person (indeed, each represen tative plaintiff had a stake exceeding the whole bill of costs), but the principle is general: The absent class members are free riders on the representative plaintiffs' legwork, and making the representatives' position financially risky would discourage class actions across the board.

It does not follow from this, however, that the prevailing defendant must bear the costs. Entrepreneurial attorneys already supply risk-bearing services in class actions. They invest legal time on contingent fee, taking the risk of failure in exchange for a premium award if the class prevails. A suit such as this, designed to generate a substantial financial return, induces lawyers to compete for the opportunity to represent the class. What we held in Rand is that, without violating ethical standards, attorneys may agree to bear the risk of a costs award, as well as the risk that their time will go uncompensated. By moving the risk of loss from the representative plaintiffs to the lawyers (who spread that risk across many cases and thus furnish a form of insurance) counsel can eliminate the financial disincentive that costs awards otherwise would create. In this case, however, the representative plaintiffs filed suit without securing from their lawyers any undertaking to pick up the tab; the lawyers have not volunteered to do so now that the plaintiffs have lost. The eight pensioners may view this as churlish--and other would-be plaintiffs may take this into account when deciding whether to sign on with these lawyers--but a decision by the representatives and their lawyers not to strike the bargain approved in Rand is a poor reason to drop the costs back in defendants' laps. If this tactic succeeded, no sane class- action lawyer would again make the promise that the plaintiffs' lawyers made in Rand.

Plaintiffs have not cited, and we have not found, any case holding that responsibility for costs must be parceled out so that no member of a class pays more than a pro rata share. A few cases hold that costs awards should be several, rather than joint, when different groups of plaintiffs raise distinct issues that give rise to segregable costs of litigation. See, e.g., In re Paoli Railroad Yard PCB Litigation, 221 F.3d 449, 468-71 (3d Cir. 2000); Davis v. Parkman, 71 F. 961, 964 (1st Cir. 1895). Plaintiffs do not contend, however, that a subset of their number was responsible for a discrete portion of defendants' costs in this case. There is no basis for an award on other than the normal joint and several terms of liability.

Affirmed.