ORDER

J.P. STADTMUELLER, District Judge.

On June 22, 2011, the Seventh Circuit affirmed this court on appeal, with the exception of the interest crediting rate methodology used to calculate lump sum payments owed to class members. In reversing and remanding that issue, the appellate court sought only to ensure that this court chose a methodology without any deference to the defendants. It further stated that, without deciding the issue, it saw no explicit problem with the methodology originally chosen.

After holding a status conference, this court ordered the parties to submit briefs laying out their positions, and permitted both parties to file a response to the others' supporting brief. With briefing completed, and the benefit of prior submissions on this issue, the court will order that defendants Retirement Plan for Employees of S.C. Johnson & Son, Inc. ("SCJ Plan") and Retirement Plan for Employees of JohnsonDiversey, Inc. ("JDI Plan") (collectively, "the Plans") apply a "twelve-year +" average in calculating the future interest crediting rates due to class members.

The issue here is the proper method used to calculate "whipsaw" payments due to eligible class members.1 In calculating a whipsaw payment, a plan participant's hypothetical account balance is first projected forward to the normal age of retirement, applying the value of future interest credit earnings. This is done by applying an interest projection rate. In the case of a variable interest rate, as here, the rate must be estimated. That value at normal retirement age is then discounted back to present value to arrive at the lump sum distribution owed to a participant that takes an early distribution. In the court's original order, it approved the use of a rolling five-year average method2 to determine a given class member's future interest crediting rate. The five-year average used the actual interest crediting rate for the year of distribution in addition to the rates for the four years immediately preceding the year of distribution. That average was then presumed to apply to each year post-distribution until the normal retirement age. In approving that methodology, the court deferred to the Plans, finding that the chosen method, with one exception remedied by the court, was a reasonable choice in selecting a methodology that would fairly estimate the future interest credits. In reversing, the Seventh Circuit sought to ensure that this court chose a methodology without deferring to the Plans, noting in dicta that it saw no inherent problems with the five-year average method that would remove it from the realm of fair estimation.

In the newly submitted briefs, the Plans re-submit their "spread" methodology for consideration on the basis of prior briefing, and then alternatively argue for the re-selection of the five-year average. The plaintiffs re-argue their stochastic method and also put forth a new methodology not previously briefed: a "20+" year long-term average approach that operates similarly to the five-year average but includes a greater amount of prior yearly data. Unlike the Plans' five-year method, the 20+ year long-term average is not a rolling average, that is, it simply continues adding yearly data without dropping older data.

First, the Plans argue that there is no requirement that the chosen methodology lead to the "best" estimate of the value of future interest credits. They are technically correct on that point; a lump-sum distribution need only include "a fair estimate of [future interest] credits." Berger v. Xerox Corp. Retirement Income Guarantee Plan, 338 F.3d 755, 761 (7th Cir. 2003). However, given the imprecise nature of estimation, not only is it likely impossible to definitively determine the "best" estimate, it is also likely that there will be more than one methodology that provides a "fair" estimate. In such situations, it is implicitly the duty of the court, if not to choose the methodology leading to the best estimate, to choose a methodology that is better at presenting a fair estimate than the competing methodologies offered by the parties.3 Accordingly, while the Seventh Circuit appeared to agree that the Plans' five-year average provided a fair estimate, the court is required to do more than simply re-affirm that the five-year average is fair and re-adopt it as a result. To do so would simply re-accord deference to the Plans.

As to the offered methods, the Plans first re-argue their previously proposed spread method, resting upon their earlier briefing. The plaintiffs likewise argue again for stochastically-derived projection rates. However, the court agrees with the Seventh Circuit that an averaging methodology is more administrable than the stochastic method, and more accurately tied to the Plans' actual interest crediting method than the spread method. Thompson v. Retirement Plan for Emps. of S.C. Johnson & Son, Inc., 651 F.3d 600, 610 n.17 (7th Cir. 2011). Moreover, the court notes that the Plans apparently agreed as well when the court initially ordered them to recalculate, considering both the spread and averaging methodologies, and selecting the five-year average. Furthermore, as the Plans have argued and was discussed in this court's prior order, the U.S. Department of the Treasury's regulations call for using an averaging methodology in projecting the value of future interest credits for nondiscrimination cross-testing, and courts have looked approvingly on application in situations such as the one at hand. Treas. Reg. § 1.401(a)(4)-8(c)(3)(v)(B); Berger, 338 F.3d at 760; Esden v. Bank of Boston, 229 F.3d 154, 170 (2d Cir. 2000). As such, the court will consider both the Plans' preferred five-year average methodology and the plaintiffs' 20+ long-term average methodology. The Plans argue that, while a five-year average method is appropriate, the plaintiffs' proposed 20+ year averaging period has no economic or legal basis.

After careful consideration, the court believes that an averaging methodology that uses an average beginning with 1986's interest rate is the most appropriate fair estimate of the future interest credits that should have been applied in valuing the plaintiffs' lump-sum distributions. The court will term this the "twelve-year +" method.4 The Treasury's nondiscrimination cross-testing regulation uses a methodology that calls for using either the current period's interest rate, or an average of the rate for the current period and one or more periods preceding, but not to exceed an aggregate five years. Treas. Reg. § 1.401(a)(4)-8(c)(3)(v)(B). The little authority there is regarding an appropriate methodology cites that regulation, but only in non-binding fashion. Berger, 338 F.3d at 760; Esden, 229 F.3d at 170. By the same token, there is no authority discussing, let alone disapproving, of using an averaging methodology that takes a greater-than-five-year span into account. The Plans claim that this court previously held that "Berger `permits an average of no more than five years,'" (Defs.' Opening Br., 7) (Docket #298), but that is disingenuous because the court only wrote that the regulation cited by the Berger court permitted no more than a five-year sample. Thompson v. Retirement Plan for Emps. of S.C. Johnson & Son, Inc., 2010 WL 3282666, at *7 (E.D. Wis. Aug. 19, 2010). Further, that decision was made in deference to the Plans, reviewing only whether the rate was a reasonable choice in complying with the law. Thus, with no case law discussing averaging periods greater than five years, the court can only divine the reason the Department of the Treasury limited the period to five years by referring to the history of the regulation's enactment, interpretation by the department, and any expert evidence.

Regarding the regulation's enactment, the original publication of the final regulation in the Federal Register contains absolutely no discussion as to why the averaging period was limited to five years. Nondiscrimination Requirements for Qualified Plans, 56 Fed. Reg. 47,524 (Sept. 19, 1991). Neither does the preamble to the subsequent revised regulations, which note only that comments to the cross-testing were still being reviewed and amendments would be proposed later. Nondiscrimination Requirements for Qualified Plans, 58 Fed. Reg. 46,773, 46,777 (Sept. 3, 1993). The subsequent Internal Revenue Service Notice that described proposed guidance and discussed the cross-testing provisions likewise did not address the five-year limit. I.R.S. Notice 96-8, 1996 WL 17901 (Feb. 5, 1996). Without the benefit of any explanation for limiting the averaging period to five years, the regulation itself is restricted to its ability to persuade that a five-year period would lead to a fair estimate, but it says very little about avoiding larger averaging periods. Pairing this with the different context of the regulation (nondiscrimination cross-testing), it is of limited instruction that the department chose not to allow averaging periods beyond five years. What's more, regulations are written with an eye toward general applicability and without the benefit of knowing the specific circumstances of any given application. Here, the court has a much more specific context in which to judge methodologies. In fact, one of the Plans' own experts previously stated that because the Plans' variable interest rate is based upon trust returns each year, as opposed to being based upon yields on Treasury securities or rates derived directly from Treasury securities, the Plans would not be eligible to use the safe harbor nondiscrimination testing that restricts interest rate averages to a five-year period. (Nicholl Report, 18) (Docket #107-6); see also Treas. Reg. § 1.401(a)(4)-8(c)(3)(iv)(B), (C)(2). While that fact does not change the court's opinion that a five-year average can lead to a fair estimate, it further reduces the regulation's persuasive authority as to the necessity of capping an average at five years. The regulations were written with respect to variable interest rates of a type that are not used here. As such, the court is not persuaded that the department's averaging period limit should be treated as an implicit judgment on the wisdom of choosing a longer period in the situation at hand.

Additionally, the Plans have submitted no expert evidence to explain the underlying economic reasons for limiting the averaging period to five years. Again, the court does not view this as a reason to doubt the fairness of using such a methodology. However, it cuts against the Plans' argument that the plaintiffs' 20+ year average has no economic basis. If the court accepted that argument, it would appear that the five-year average similarly lacks any economic basis, being grounded only in persuasive legal authority. Instead, a basic familiarity with math reveals that an average created with fewer data points will generally be susceptible to skewing by the existence of a single data point that constitutes a considerably outlying value. But an average that contains more data points tends to smooth out the effect of individual points with large variations in value. In fact, the Plans, in prior briefing, argued forcefully that it would be unfair to use a method looking only to one data point because it would lead to wide variation (or "extreme inequities") between additional payments made to class members after recalculation. (Defs.' July 12, 2010 Joint Supplemental Memo., 11-12) (Docket #242).5 The Plans' logical conclusion was that a five-year average more fairly estimated the value of future interest credits, thus illustrating that an average containing five data points better controls for outlying data points as compared to simply using a single data point.6 As such, assuming the goal is to avoid overly disparate treatment of class members, including more data points in an average is likely to lead to a fairer estimate of future interest credits. With that in mind, the court recognizes that there are other factors that might bear upon the fairness of an estimate when determining which data points ought to be included in any average; but it is relatively beyond dispute that including more data points generally presents a more accurate average.

The Plans also argue that the plaintiffs present "no economically sound basis" for selecting the 20+ year average because there is no explanation as to why six, seven, eight, thirty, or even fifty years of data is not appropriate. But likewise, the Plans provide no reason why five years is the proper ceiling, other than an appeal to persuasive law: a regulation of little persuasive value beyond its endorsement of averaging in general; and two favorable appellate citations that do not analyze the issue. The court is unconvinced that a lack of expert evidence forecloses a determination that an averaging methodology using more than five years of data will lead to a fair estimate.

Thus, the court proceeds to determine which data points ought to be part of the average, recognizing that the inclusion of more data will result in a fairer estimate assuming there is no other persuasive reason to exclude a particular data point. As to that issue, the Plans note that their expert, Dr. Cathy Niden, used a twelve-year period in developing her average "spread" rate. Dr. Niden used a twelve-year period because the investment mix for assets of the SCJ Plan prior to 1986 were arguably not comparable to the mix in place in 1998 when the SCJ Plan was first converted to its current form. (Niden Report ¶¶ 23-26 & Ex. D) (Docket #175-1). As Dr. Niden explained in her expert report, the mix of assets was important because "expected future returns are related to the risk of the assets that compose the investments." (Niden Report ¶ 23). This reasoning counsels against including the full twenty-plus years of data that the plaintiffs advocate.

However, the court is convinced that something more than a five-year average should be used. Particularly salient is the fact that the actual interest crediting rates for 2000, 2001, and 2002 show a fairly anomalous return of 4% for each year, the lowest possible rate under the plan terms. (Aug. 19, 2010 Order, 6) (Docket #246). While a return of 4% is not itself strange, three consecutive years with those results is the worst such period of returns as far back as there is data. (Aug. 19, 2010 Order, 6); (Lowman Decl. ¶ 8) (Docket #240). As a result, a rolling five-year average results in an estimated future interest crediting rate of 8.142% for SCJ Plan members that took a distribution in 2004, yet a member that took a distribution a mere two years later in 2006 would receive a 9.798% rate, or a 10.726% rate for a member that took a distribution in 2001. While a five-year average may generally satisfy the fairness threshold, in this instance it leads to impressive variation because of the existence of a three-year bad spell. While the nature of determining damages tasks the court with choosing a methodology that will fairly estimate future interest credit earnings on the basis of what should have been paid, and thus without reliance on the hindsight of future interest rates' actual values, it remains the case that members taking distributions in a relatively similar time frame could logically expect to receive estimated future interest rates that are relatively similar. For example, a member with 25 years remaining until retirement that took a distribution in 2001, and a member with 22 years remaining until retirement that took a distribution in 2004 are both being credited with an estimated interest rate for each year from 2005 until 2026. Had both members left the money in their accounts until 2026, they would receive the same actual interest rate for each year. Therefore, it is reasonable to expect that an estimated rate would fall within a narrow window of variation given that they took their actual distributions within a relatively narrow window of years of each other. A five-year average, in this particular case, does not accomplish that. On the other hand, a twelve-year + average does accomplish this. The rates for class members range from 9.41% to 9.93% using the twelve-year + average method, unlike the rolling five-year average's spread from 8.142% to 10.726%. Moreover, taking data only from 1986 on takes into consideration the Plans' expert's concerns over investment mix and its effect on returns. Finally, the method's non-rolling quality avoids the arbitrary ignoring of data points as between class members.

As such, the court views a twelve-year + average as leading to a fair estimate of future interest crediting rates. The average should be calculated using the rates from 1986 through the year of distribution for any given class member.

Accordingly,

IT IS ORDERED that the defendants shall apply a "twelve-year +" average of crediting rates in determining the value of underpayment made to SCJ Lump Sum Subclass A and JDI Lump Sum Subclass A members. In constructing the average, the defendants shall use the interest crediting rates from 1986 through the year of distribution for a given class member. Defendant Retirement Plan for Employees of JohnsonDiversey, Inc., in constructing its average of crediting rates, shall use the interest crediting rate for the Retirement Plan for Employees of S.C. Johnson & Son, Inc. for 1998, as previously ordered, as well as for prior years.

The clerk of court is directed to amend the November 18, 2010 Amended Judgment (Docket #264) accordingly.