OPINION OF THE COURT
McKEE, Circuit Judge.
This tax dispute arose in the course of bankruptcy proceedings for Harvard Industries, Inc. and related entities (collectively “Harvard”). Its resolution requires us to determine the meaning of “specified liability losses” as that phrase is used in 26 U.S.C. § 172(f).1 In the bankruptcy court, Harvard attempted to collect a federal tax refund for the 1986 tax year based upon three categories of “specified liability losses” incurred in the 1996 tax year that purportedly qualified for a special ten-year net operating loss carry-back pursuant to 26 U.S.C. § 172(b)(1)(C).2 The bankrupt[446]*446cy court allowed the carry-back for each of the claimed expenses over the government’s objection.
On appeal from the bankruptcy court, the district court ruled that (i) payments to a qualified pension plan that were made pursuant to a settlement agreement with the Pension Benefit Guaranty Corporation (“PBGC”)3 did not “arise under Federal ... law” and were therefore not “specified liability losses” under § 172; (ii) losses incurred in relation to the manufacture of defective lock nuts were not “product liability” damages within the meaning of § 172, and hence were not “specified liability losses”; but (iii) payments made pursuant to a retrospective workers’ compensation insurance policy were properly deductible as “specified liability losses” in the 1996 tax year. For the reasons that follow, we will affirm in part and reverse in part.
I. FACTUAL BACKGROUND
In 1986, Harvard earned profits of approximately $6.5 million and hence paid a total of $2,442,175 in federal income taxes. For the 1996 tax year, Harvard sustained a net operating loss of $41,399,563. On April 23, 1998, Harvard filed an Amended Corporation Income Tax Return in which it claimed a refund in the amount of $2,435,872 for the 1986 tax year based on a specified liability loss generated during the 1996 tax year that was purportedly eligible to be carried back ten years pursuant to § 172 of the Internal Revenue Code (“I.R.C.”).4 In a Notice of Partial Claim Allowance dated February 23, 1999, the Internal Revenue Service (“IRS”) allowed the carry-back for such losses as were attributable to Workers’ Compensation payments, but denied the remainder of the claimed refund.
On March 5, 1999, Harvard filed a protest to the First Partial Disallowance and challenged the Service’s basis for denying portions of its refund claim. Harvard also expanded the refund claim to include, among other things: (i) “product liability” payments in the amount of $3,829,807 which included forgiven accounts receivable and a settlement payment to one distributor in connection with defective lock-nuts manufactured by a Harvard operating division; and (ii) prior year pension payments in the amount of $6,000,000 (the “PBGC Payments”).
The IRS issued a Second Notice of Partial Disallowance on October 1, 1999, denying Harvard’s refund application for the lock-nuts and the PBGC payments. The IRS denied the claim related to the lock-nuts because: (i) Harvard’s liability was based on breach of contract and breach of implied warranty of merchantability, as opposed to product liability; (ii) Harvard’s customers suffered no physical or emotional harm due to the defective lock-nuts; (iii) [447]*447costs incurred by Harvard were for repair and replacement of the lock-nuts. The IRS also determined that the pension payments were not eligible for a ten-year carry-back because the Code requires that the event giving rise to an eligible liability “under state or federal law” must occur at least three years before the tax year in question, 1996 in this case. The IRS’s position was that because the payments were made pursuant to a settlement agreement with the PBGC in 1994, they did not meet the Code’s requirements for eligible “special liability loss” carry-backs. The IRS also took the position that the formation of pension plans and the decision to enter a Settlement Agreement with the PBGC regarding additional funding requirements for the pension plans were voluntary decisions of Harvard, not “arising under federal law” as required by the Code. Rather, they “related to” an obligation under federal law, which is not enough to satisfy the “arising under” element required for the special carry-back provision of the Code. As we explain below, this ongoing dispute was ultimately brought before the bankruptcy court.
A. Losses Related to Defective Lock-Nuts
Elastic Stop Nut of America (“ESNA”), an operating division of Harvard, manufactured lock-nuts for use in commercial and military aircraft engines and airframes. Harvard sold the lock-nuts to various distributors, who resold them to aircraft manufacturers. Military and commercial specifications required that the lock-nuts be baked for 23 hours in order to withstand extreme temperatures during use. Failure to comply with this requirement could result in a condition called “hydrogen embrittlement” which could cause the lock-nuts to crack and fail.
In 1993, it was discovered in the course of an internal investigation that certain of Harvard’s lock-nuts were defective because they had not been baked for 23 hours as required. When the defect was discovered, Harvard informed its customers and stopped shipping the lock-nuts pending further investigation. Prior to the time Harvard stopped shipping the lock-nuts, there were no reported instances in which the failure of an ESNA lock-nut caused an accident or resulted in personal injury or property damage. In some cases, efforts were made to recall and rework some of the lock-nuts. However, several distributors who had received the defective lock-nuts refused to pay for them because they could not be resold and/or had to be recalled.
Harvard’s largest customer — Harco— filed suit against Harvard based on the sale of defective lock-nuts, alleging: (1) breach of contract; (2) breach of implied warranty of merchantability; (3) breach of the implied covenant of good faith and fair dealing; (4) fraud; (5) negligent misrepresentation; and (6) civil RICO violations. The suit was ultimately settled in an agreement dated April 22, 1996. Pursuant to that agreement, Harvard paid Harco $820,000 and Harco agreed to “release and discharge” Harvard from “any and all claims asserted” in Harco’s complaint. Harvard then entered into settlement agreements with other distributors, whereby ESNA forgave a total of $3,009,807 in receivables for the lock-nuts. Harvard subsequently claimed that it should be allowed to treat all these expenses as “product liability” losses eligible for a ten-year carry-back.
B. PBGC Settlement Pension Payments
Harvard filed for Chapter 11 bankruptcy in May of 1991.5 Thereafter, the bank[448]*448ruptcy court confirmed a plan of reorganization which required Harvard to pay all holders of allowed unsecured claims 100 cents on the dollar by 1994. In order to meet its obligations under the plan, Harvard sought to obtain $100 million in financing by offering senior unsecured notes.
However, the PBGC was concerned about the issuance of the notes. Harvard’s pension plans had a “substantial amount of unfunded benefit liabilities” due to erroneous actuarial assumptions underlying pension plan contributions for 1992 and 1993. The PBGC therefore took the position that the note offering might provide “sufficient basis for the PBGC to seek termination of one or more of [Harvard’s] pension plans pursuant to section 4042(a)(4) of ERISA, [29 U.S.C. § 1342(a)(4) ].” Negotiations followed in which the PBGC and Harvard reached a settlement agreement. Pursuant to that agreement, Harvard made a $ 6 million additional contribution to its pension plans in 1996 and agreed to pay an additional $1.5 million for each of twelve consecutive quarters thereafter.6
The PBGC Settlement Agreement contains restrictions on the amount and use of the proposed Senior Notes. Harvard warranted in the agreement that: “as of the date of execution of this Settlement Agreement there are no past due minimum funding contributions owed to any of’ its pension plans, and the PBGC agreed that it would not institute proceedings to terminate any of taxpayer’s pension plans “as a result of the Senior Notes offering.”
C. Workers’ Compensation Payments
From April 1988 to April 1992, Harvard annually purchased insurance policies from the Wausau Insurance Company (“Wausau”). The policies included insurance for general liability, workers’ compensation and automobile insurance. Harvard describes the Wausau workers’ compensation polices as “retrospective insurance plans.” Pursuant to these policies, Harvard paid an initial premium at the beginning of each policy year based on actuarial assumptions about the amount of claims that would be paid. Once Harvard paid its premium to Wausau, Wausau had access to these funds and used them to pay claims covered by the policy.
At the end of each policy year, adjustments were made to the premium based on the difference between the actual amount paid out on claims and the expected claim amounts that had been estimated based on actuarial assumptions. Even after the policy years expired, as claims arising in those years were paid, Harvard could be required to submit additional payments. Wausau periodically sent reports to Harvard concerning claim activity. By comparing year to year reports, Wausau could determine if Harvard had to make additional payments to cover any shortfall for each plan year. Harvard also paid taxes and premium expenses for plan administration, calculated as a percentage of claims expenses.
According to testimony given by a Harvard representative, Harvard’s records indicated that the retrospective adjustments pertaining to the refund request at issue here were sent to Harvard around October 1995. Wausau and Harvard then commenced negotiations and ultimately came to an agreement as to the appropriate adjustments in early 1996. The Trust also seeks to carry-back those payments to Wausau as “specified liability losses” arising under state law because they con[449]*449stitute Harvard’s obligation to provide workers’ compensation benefits for its employees.
II. PROCEDURAL BACKGROUND.
On January 15, 2002, Harvard, along with several subsidiaries, filed a voluntary petition for bankruptcy under Chapter 11 of the Bankruptcy Code. Thereafter, Harvard filed a “Motion Requesting a Determination as to Debtor’s Rights to a Tax Refund.” The substance of the motion dealt with the three categories of supposed “specified liability losses” described above. Harvard argued that each expense qualified for a refund of federal taxes paid for the 1986 tax year. The motion was heard as a contested matter pursuant to Fed. R. Bank. P. 9014. While the motion was pending, Harvard’s Reorganization Plan was confirmed and the Harvard Secured Creditors Liquidation Trust (the “Trust”) became the party in interest with respect to any potential refund. Eventually, the Trust and the government filed cross-motions for summary judgment in the bankruptcy court.
On March 24, 2005, the bankruptcy court granted the Trust’s summary judgment motion with respect to two of the three categories of specified liability loss expenses at issue. In re Harvard Indus., Inc., 324 B.R. 238 (Bankr.D.N.J.2005). The court ruled that the lock-nut related payments were product liability losses as they were a “liability of the taxpayer for damages on account of ... loss of the use of property” in accordance with I.R.C. § 171(f)(4). The court reasoned that “[sjince the term ‘property’ is not defined in the statute,” it must be accorded “its ordinary meaning ... [which] would include the Lock-Nuts at issue here.” Id. at 241. The bankruptcy court also ruled that the pension payments “arose under ERISA,” and were therefore also specified liability losses under the Tax Code. Id. at 242. Thus, Harvard was entitled to a refund as a result of the allowance of these expenses. The bankruptcy court denied Harvard’s motion with respect to the third category of claimed expenses (workmen compensation premiums) pending additional discovery, and denied the government’s cross-motion for summary judgment. The amount of the overpayment ordered by the bankruptcy court exceeded the 1986 tax paid when added to the refund amounts already received by Harvard. Therefore, no further refunds could be ordered and the bankruptcy court’s summary judgment order was final. Thereafter, the government appealed to the United States District Court for the District of New Jersey, which had jurisdiction under 28 U.S.C. § 158(a).
For reasons we explain below, the district court reversed the bankruptcy court’s order with regard to the lock-nut expenses and the PBGC payments, and remanded the matter for resolution of the third disputed category of losses. After a hearing, the bankruptcy court ordered that Harvard was entitled to a refund with respect to the retrospective adjustments to its workers compensation plan, but held that administrative fees associated with the plan could not be included in the carry-back. On November 22, 2006, the Trust appealed to the district court and the district court subsequently affirmed in part, reversed with respect to administrative costs associated with the policy, and remanded to the bankruptcy court for entry of judgment. The district court’s order is a final order because it disposes of all claims with respect to all parties. Thereafter, the Trust appealed to this court. We have jurisdiction pursuant to 28 U.S.C. § 158(d).
III. STANDARD OF REVIEW
Our standard of review is the same as the district court’s review of the bankrupt[450]*450cy court’s ruling. In re Schick, 418 F.3d 321, 323 (3d Cir.2005). We review an order granting summary judgment de novo. American Flint Glass Workers Union v. Anchor Resolution Corp., 197 F.3d 76, 80 (3d Cir.1999). The bankruptcy court’s application of the “all events” test is also reviewed de novo.7 ABKCO Indus., Inc. v. Commissioner, 482 F.2d 150, 154 (3d Cir. 1973). Factual findings are reviewed for clear error. Nantucket Investors. II v. California Fed. Bank, 61 F.3d 197, 203 (3d Cir.1995).
IV. ANALYSIS.
All of the issues before us turn on the interpretation of § 172 of the I.R.C. In 1996, that section allowed for certain portions of net operating losses, called “specified liability losses,” to be carried back ten years to offset tax liabilities incurred in more profitable years.
During the period in question, I.R.C. § 172(f) defined “specified liability loss” as follows:
(1) In General. — The term “specified liability loss” means the sum of the following amounts to the extent taken into account in computing the net operating loss for the taxable year:
(A) Any amount allowable as a deduction under section 162 or 165 which is attributable to—
(i) product liability, or
(ii) expenses incurred in the investigation or settlement of, or opposition to, claims against the taxpayer on account of product liability.
(B) Any amount (not described in subparagraph (A)) allowable as a deduction under this chapter with respect to a liability which arises under Federal or State law, or out of any tort of the taxpayer if—
(i) in the case of a liability arising out of a Federal or State law, the act (or failure to act) giving rise to such liability occurs at least 3 years before the beginning of the taxable year 8
The Trust contends that all three categories of 1996 expenses at issue here qualify as “specified liability losses” under this section of the Code and can thus be carried back to 1986 — making Harvard eligible for a refund from that year. As noted earlier, Harvard claims that expenses related to the lock-nuts qualify as “product liability losses,” while the pension payments and the workers’ compensation insurance payments purportedly “arise out of a Federal or State law,” and therefore satisfy the requirements of § 172(f).
As there is no binding authority interpreting this statute, we rely on basic [451]*451tenets of statutory interpretation. When interpreting a statute, “the literal meaning of the statute is most important, and we are always to read the statute in its ordinary and natural sense.” Galloway v. United States, 492 F.3d 219, 223 (3d Cir. 2007) (internal quotation marks and citations omitted). In construing the Tax Code, we “strictly construe deductions and allow such deductions only as there is a clear provision therefor.” Id. Moreover, we rely on the legislative history only where the text itself is ambiguous. Id. We have recently ruled that where a provision of the I.R.C. is ambiguous, we apply a Chevron analysis to any applicable treasury regulations. Swallows Holding, Ltd. v. Comm’r, 515 F.3d 162 (3d Cir.2008).
A. “Product Liability” Losses.
The I.R.C. defines “product liability” for purposes of section 172(f)(1)(A)(I) as:
(A) liability of the taxpayer for damages on account of physical injury or emotional harm to individuals or damage to or loss of the use of property, on account of any defect in any product which is manufactured, leased, or sold by the taxpayer, but only if
(B) such injury, harm or damage arises after the taxpayer has completed or terminated operations with respect to, and has relinquished possession of, such product.
26 U.S.C. § 172(f)(4). Neither the Supreme Court, nor any circuit court of appeals has interpreted this provision.
The bankruptcy court analyzed the language of the statute and concluded that Harvard’s settlement with Harco and other customers qualified as “liability ... for damages on account of ... loss of the use of property.” 324 B.R. at 241 (quoting I.R.C. § 172(f)(4)). It reasoned that the word “property,” which is not defined in the statute, should be read to include the lock-nuts themselves. Therefore, the court concluded:
Loss of the use of the defective property is precisely what occurred here. Harvard’s customers were distributors who were unable to use the Lock-Nuts manufactured by [Harvard] because of a defect known as hydrogen embrittlement. Here again, the court gives the term “use” its plain meaning which would include intended use as an item to resell.
Id.
The district court rejected the interpretation of the bankruptcy court. It reasoned that:
Loss contemplates possession followed by the failure to maintain possession. Harvard’s customers did not have possession of lock-nuts fit for resale at any point; they merely had possession of defective lock-nuts that were unfit for resale. Consequently, Harvard’s customers could not have lost the use of the property for its intended purpose where they did not possess usable lock-nuts in the first place.
Additionally, Section 172(f)(4)(B) requires that “such injury, harm, or damage arises after the taxpayer has completed or terminated operations with respect to, and has relinquished possession of, such product.” In the instant case, the defect that gave rise to Harvard’s liability arose during the manufacturing of the lock-nuts, as Harvard’s own brief admits. [ ] Since the damage to the property clearly occurred before Harvard relinquished possession of the product, the damage to the lock-nuts is excepted from the statutory definition of product liability as stated in 26 U.S.C. § 172(f)(4).
App. 38.
As we have just noted, product liability is “liability of the taxpayer for damages on [452]*452account of physical injury or emotional harm to individuals or damage to or loss of the use of property, on account of any defect in any product which is manufactured, leased, or sold by the taxpayer.... ” I.R.C. § 172(f)(4)(A) (emphasis added). It is uncontested that the lock-nuts were defective. It is also uncontested that none of Harvard’s customers suffered physical injury or emotional harm because of the defective lock-nuts.9 There is also no allegation that any of the lock-nuts caused any damage to other property of any customer or “down-stream” user. (For instance, had a defective lock-nut caused a plane to crash, Harvard might well have been liable for the cost of replacing the plane as well as other damages.) The question then is whether the distributor’s inability to resell the defective product itself qualifies as “damage to or loss of the use of property.”
Both the district court and the bankruptcy court examined the statute closely, referencing dictionary meanings for each significant term. Yet, those two courts arrived at opposite conclusions. This clearly suggests an ambiguity in the language of the statute. Much of the textual ambiguity arises from the fact that it is not clear whether Congress intended “property” in the phrase, “loss of the use of property,” to include the defective product itself as opposed to the property of downstream purchasers or users to which the defective product has caused loss or damage.
In arguing that “property” refers to something other than the actual lock-nuts, the government focuses on the fact that Congress used “property” and “product” differently in the statute. Relying on this distinction, the government reminds us that:
Where the statutory language refers to the defective product, it uses the term “product,” which term appears once in subparagraph (A) and once in subparagraph (B). I.R.C. § 172(f)(4). Subparagraph (A) refers to “damages ... on account of any defect in any product,” while subparagraph (B) refers to the requirement that the “damage arises after the taxpayer has completed or terminated operations with respect to, and has relinquished possession of, such product.” I.R.C. § 172(f)(4). Not only is the defective product referred to in both instances by the term “product,” but the second instance in effect refers back to the first instance by using the term “such product.” Id.
On the other hand, the term “product” does not appear in the phrase “loss of the use of property,” ie., the statutory language does not refer to a “loss of use of the product or other property.” Id. Rather, the statute refers to a “loss of use of property.” Id. In this context, the term “product,” and not the term “property,” refers to the lock-nuts. Thus, when viewed in the context of the definition as a whole, the distributor’s inability to resell the lock-nuts did not constitute a loss of use of property.
Reply Br. 35-36. Although this approach has some surface appeal, we believe it actually does more to demonstrate the difficulty of textual analysis than to establish the congressional intent underlying the language we must interpret.
We therefore turn to legislative history for guidance. In re Unisys Sav. Plan Litigation, 74 F.3d 420, 444 (3d Cir.1996) [453]*453(“If the statutory language is unclear, we then look to [a statute’s] legislative history.”). The House Conference Report stated that “[t]he definition of product liability under the senate amendment is intended to include the kinds of damages that are recoverable under prevalent theories of product liability.” H.R.Rep. No. 95-1800, at 286 (1978), U.S.Code Cong. & Admin.News 1978, pp. 7198, 7281. It went on to state that “[t]he definition of product liability in the amendment does not include liabilities arising under warranty, which essentially are contract liabilities.” Id. at 287, U.S.Code Cong. & Admin.News 1978, pp. 7198, 7282.10 Unfortunately, there was more than one prevalent theory about the kinds of damages recoverable under product liability law when the statute was enacted. Fortunately, the Supreme Court has discussed the divergent views of product liability that were viable around that time.
In East River Steamship Corp. v. Transamerica Delaval, Inc., 476 U.S. 858, 867-70, 106 S.Ct. 2295, 90 L.Ed.2d 865 (1986), a defectively designed ship turbine component malfunctioned and damaged the turbine itself without harming any other part of the ship. The Supreme Court was called upon to determine whether, in the context of admiralty law, injury to a product itself was the kind of harm that should be addressed by contract law or product liability law. This was then a question of first impression in admiralty. The Court began its analysis by noting that “general maritime law is an amalgam of traditional common-law rules, modifications of those rules, and newly created rules,” which are “[d]rawn from state and federal sources.” Id. at 864-65, 106 S.Ct. 2295. It is this analysis of prevailing common law rules that makes the Court’s opinion useful to our analysis here.
The Court viewed the “paradigmatic products-liability action [as] one where a product ‘reasonably certain to place life and limb in peril,’ distributed without reinspection, causes bodily injury.” Id. at 866, 106 S.Ct. 2295 (citing MacPherson v. Buick Motor Co., 217 N.Y. 382, 111 N.E. 1050 (1916)). It then discussed the expansion of products liability to include protection against property damage based on similar concerns of safety. Id. at 867, 106 S.Ct. 2295. However, the expansion traditionally only involved cases where “the defective product damages other property.” Id. (emphasis added).
The Court described the “majority approach” as one which held that there should be no action in tort for “purely monetary harm” in order to “preserv[e] a proper role for the law of warranty....” Id. at 868, 106 S.Ct. 2295 (citation omitted). The minority approach “held that a manufacturer’s duty to make nondefective products encompassed injury to the product itself, whether or not the defect created an unreasonable risk of harm.” Id. at 868-69, 106 S.Ct. 2295. After evaluating the merits of these different approaches, [454]*454the Court concluded that where the only-injury was to the product itself, “the resulting loss due to repair costs, decreased value, and lost profits is essentially the failure of the purchaser to receive the benefit of its bargain — traditionally the core concern of contract law.” Id. at 869, 106 S.Ct. 2295 (citing E. Farnsworth, Contracts § 12.8, pp. 839-40 (1982)). The Court concluded that “a manufacturer in a commercial relationship has no duty under either a negligence or strict produets-liability theory to prevent a product from injuring itself.” Id. The Court also reasoned that the policy concerns for public safety were not as compelling in these circumstances as in those where bodily injury or harm to other property occurred.
Thereafter, we had to decide what rule Pennsylvania would adopt when a defective product “damaged itself.” In Aloe Coal Co. v. Clark Equipment Co., 816 F.2d 110 (3d Cir.1987), we predicted that “Pennsylvania courts, although not bound to do so, would nevertheless adopt the Supreme Court’s reasoning in East River.” Id. at 112. In so doing, we reversed a conclusion we had reached in a previous case, because we were persuaded by the “cogent reasoning” of East River. Id. at 119.
Neither of these cases controls our present inquiry under the Tax Code. However, we continue to find the reasoning of East River persuasive, and the distinction it draws between warranty and contract damages on the one hand, and product liability and tort damages on the other, is similar to that drawn by the Congress that drafted and enacted 26 U.S.C. § 172(f). Moreover, this approach is reinforced here because Harco sued Harvard on various theories of contract and warranty liability based on the defective lock-nuts. Harco did not assert a product liability cause of action. Thus, the damages here are “liabilities arising under warranty,” which Congress did not intend to include in the statute.
As noted earlier, the taxpayer has the burden of proving its eligibility for a deduction, and statutes authorizing deductions are a matter of legislative grace and are to be construed narrowly unless the text of the statute authorizing the deduction reflects a different congressional intent. See B.A. Properties Inc., v. Government of the Virgin Islands, 299 F.3d 207 (3d Cir.2002). Viewed in that context, we are not persuaded by the Trust’s argument that the IRS’s interpretation of the statute will leave manufacturers with no incentive to make safe products. In fact, the argument is specious. Even if corporations are not allowed to carry-back this deduction 10 years — they may still take a deduction for such expenses in the applicable tax year. Furthermore, regardless of how such liabilities are treated for tax purposes, the threat of products liability and other claims hangs over a company that makes unsafe products. Here, for example, the potential products liability and tort recovery from death and injury that could have resulted from a plane crash caused by the defective lock-nuts would dwarf the claimed tax benefit of allowing a ten year carry-back.
We therefore conclude that the district court did not err in reversing the bankruptcy court’s conclusion that the loss fell within the scope of § 172(f). The district court was correct in accepting the government’s position and disallowing the Trust’s claim that the payments for defective lock-nuts qualified for the ten year carry-back.
B. PBGC Payments
A taxpayer may also claim a specified liability loss if the deduction “arises under a Federal or State law” if “the act (or failure to act) giving rise to such liability occurs at least 3 years before the begin[455]*455ning of the taxable year” and “the taxpayer used an accrual method of accounting throughout the period or periods during which the acts or failures to act giving rise to such liability occurred.”11 I.R.C. § 172(f)(1)(B).
As explained above, Harvard made $6 million in payments to its various pension plans in the 1996 tax year pursuant to the settlement agreement with the PBGC. The Trust argues that these payments “arose under” the Employee Retirement Income Security Act, 29 U.S.C. § 1001 et. seq. (“ERISA”). ERISA sets minimum standards for most voluntarily established pension and health plans in private industry. The Trust maintains that the underfunding of Harvard’s pension plans in 1992 and 1993 created ERISA liability and that the liability therefore arose under ERISA and was partially discharged through the 1996 PBGC payments. Thus, according to the Trust, those payments meet the statute’s requirements because they constitute a liability that arose under federal law and accrued at least three years before the loss.
The government argues that these payments are not a specialized liability loss for two reasons. First, the PBGC payments are not rooted in federal law; rather, they resulted from choices made by Harvard. Second, the relevant act was the choice to enter into a settlement agreement with the PBGC in 1994 — an act which occurred less than three years before the payments.
However, we are persuaded by the bankruptcy court’s insightful analysis of this issue. We therefore adopt the bankruptcy court’s cogent and persuasive discussion of this issue:
In arguing that the payments did not arise under federal law, the IRS focuses on the fact that Harvard had satisfied its minimum funding requirements under section 412 of the IRC. That argument ignores the fact that those are not the only payment obligations under ERISA. Additional funding requirements may be triggered by a plan’s unfunded current liability. [ ]. That was the case here because the PBGC had determined that Harvard had unfunded current liabilities in the tax years 1992 and 1993. It is certainly true that Harvard’s settlement with the PBGC on that issue was motivated by its desire to issue senior notes to fund its plan of reorganization without objection from the PBGC, but that does not change the ultimate fact that the plans had unfunded liabilities in 1992 and 1993. Thus, to maintain its qualified status Harvard was required by law to make those payments.
That, of course, leads to the IRS’s other argument: that the need for additional contributions to the pension plans arose out of a choice made by Harvard to maintain qualified pension plans for its employees. In a recent decision on this issue the Federal Circuit Court of Appeals stated that “the nature and amount of the liability must be traceable to a specific law and cannot be the result of choices made by the taxpayer or others.” Major Paint Co. v. United States, 334 F.3d 1042 (Fed.Cir. 2003). While that decision is interesting, it does not instruct a court on where it must draw the line regarding what constitutes a choice made by a taxpayer. At some level everything involves a choice. It is frequently recognized that liability for workers’ compensation claims may qualify for specified liability loss status, Host Mar[456]*456riott v. United States, 113 F.Supp.2d 790 (D.Md.2000), aff'd 267 F.3d 363 (4th Cir. 2001), yet that liability only arises because an employer makes the decision to hire workers who are covered by that law. A similarly slippery slope is apparent here. While it is certainly true that offering an ERISA qualified pension plan to its employees was a voluntary business decision by Harvard, the court finds that the more prudent interpretation would be to find that once a decision like that is made then Harvard was bound by all of ERISA’s regulations. Thus, complying with ERISA’s funding requirements was not a voluntary decision on the part of Harvard, it was required by federal law.
The next issue is whether the liability arose within three years prior to the beginning of the taxable year at issue. The IRS takes the position that the final act fixing Harvard’s liability occurred on July 26, 1994, the date Harvard entered into its agreement with the PBGC. The court finds that argument to be misplaced. The agreement with the PBGC did nothing to create Harvard’s liability, it was merely the settlement of how that liability would be paid. The liability itself was created in tax years 1992 and 1993 due to Harvard’s reliance on inaccurate actuarial assumptions. [ ]. Therefore, the court finds that the liability arose more than three years prior to the relevant tax year. Accordingly, the court will grant summary judgment in favor of Harvard on the issue of its pension plan payments qualifying as specified liability losses.
324 B.R. at 242-43.
As is evident from the portion of the bankruptcy court’s opinion set forth above, that court’s analysis was guided by the decision in Major Paint Co. v. United States, 334 F.3d 1042 (Fed.Cir.2003). There, the court held that in order for a liability to “arise under” a federal law, “the nature and the amount of the liability must be traceable to a specific law and cannot be the result of choices made by the taxpayer and others.” Id., at 1046. Major Paint is the latest of only four cases that have addressed the meaning of “arising under” in § 172.
In Sealy Corporation v. Commissioner, a taxpayer argued that professional fees the company paid to have filings required by the SEC and ERISA prepared, as well as costs incurred during an IRS audit, “arose under federal law” and should therefore qualify for the ten-year carry-back. 171 F.3d 655, 656 (9th Cir.1999). The Court of Appeals rejected this argument, holding that “[t]he act giving rise to each of the liabilities in question was the contractual act by which Sealy engaged lawyers or accountants” and that these acts “did not occur at least three years before [the tax year in question].” Id. at 657.
In Host Marriott Corp v. United States, the Court of Appeals for the Fourth Circuit adopted the reasoning of the district court in holding that interest on a federal income tax deficiency was a specified liability loss. 267 F.3d 363, 365 (4th Cir.2001). The district court had noted that “[t]he liability for federal income tax deficiency interest arises out of 26 U.S.C. § 6601(a) under a rate established by § 6621.” Host Marriott Corporation v. United States, 113 F.Supp.2d 790, 793 (D.Md.2000). The district court also distinguished Sealy, by noting that the taxpayer’s liability for tax deficiency interest is “set by federal ... law, not by [taxpayer’s] choice.” Id. at 794.
Finally, in Intermet Corporation v. Commissioner, the tax court held that state tax deficiencies and interest on federal and state tax deficiencies are specified liability losses because federal law “ex[457]*457pressly imposes” those liabilities.” 117 T.C. 133, 140, 2001 WL 1164198 (2001).
As the bankruptcy court mentioned, in Major Paint, the Court of Appeals for the Federal Circuit had to decide whether fees paid to various professionals employed to assist the taxpayer during bankruptcy proceedings “arose under federal law.” The taxpayer argued that the costs arose under the Bankruptcy Code and emphasized that a bankruptcy judge, rather than a contract, determines when and how outside professionals will be paid. Id. at 1046. The court conceded that “[t]he Bankruptcy Code does require the appointment of a committee of creditors holding unsecured claims” and the Code further provides that the committee “may select and authorize the employment ... of one or more attorneys, accountants, or other agents, to represent or perform services for such committee.” However, the court in Major Paint reasoned that “to say that simply because an entity files for bankruptcy any costs for outside professionals “arise under” the bankruptcy code in the context of I.R.C. § 172(f) stretches the limits of the Tax Code.” Id. The court in Major Paint agreed with the Sealy court that it is the act that immediately gives rise to the liability that must arise out of federal law, and not a “chain of causes” that can be traced to a federal law no matter how attenuated or nuanced the link. Id. at 1047.
Although specified liability losses must obviously “arise under” federal law rather than merely be “related to” it, we believe that formulation is, by itself, too simplistic to be determinative here because it merely states a conclusion based on reiterating the statutory text. That, without more, does not create a useful framework for analyzing a particular expense or the specific expenditure at issue here. We think the link between payments made pursuant to a settlement agreement with a federal agency threatening enforcement action and the underlying federal obligation to comply with ERISA is sufficiently direct that it may be said to “arise under federal law” and therefore qualify for the ten-year carry-back.
We also agree that the liability itself arose in 1992 and 1993, and the 1994 agreement was merely the mechanism for discharging that liability. Moreover, just as the payments under the settlement agreement are so directly related to the ERISA liability as to have arisen under ERISA, we also conclude that the payments must be treated as arising in 1992 and 1993 when the underlying liability arose, and not when the agreement enforcing that liability was executed. The date of the latter has nothing to do with the fact that the liabilities would have existed in 1992 and 1993 whether or not the 1994 agreement was ever entered into. Accordingly, payments made pursuant to that 1994 agreement were entitled to the ten year carry-back under § 172.
The bankruptcy court concluded that the Trust had proven that Harvard incurred $2,076,066 in workers’ compensation and product liability expenses in the years in question. The government argues that was error because all premiums for insurance were paid from April 1988 to April 1991, and there was no evidence of payments in the 1996 tax year.
The bankruptcy court found that Harvard owned retrospective insurance workers’ compensation and general liability policies from Wausau for four years, April 15, 1988 to April 30, 1992. Yearly premiums were based on Harvard’s actual loss experience during the applicable term. The policies required prepayment of an initial premium in the policy year. That pay[458]*458ment was intended as both a premium for traditional insurance and a prepayment designed to cover anticipated claims under the policies. Even after the policy years expired, Harvard was required to make further premium payments to Wausau as claims arising in those years were paid. Until the retrospective premium was finally agreed upon, Harvard could not determine what amount to pay Wausau.
Based on the language of I.R.C. § 172, the bankruptcy court and the district court both concluded that retrospective premium adjustments were properly considered specified liability losses, as they arose under state workers’ compensation requirements. As we find no clear error in the bankruptcy court’s factual findings, we will therefore affirm its findings regarding when the payments were made, and for what purpose.
The district court concluded that the bankruptcy court erred in excluding the “administrative expense component” of the policies from the ten-year carry-back. The district court noted that Harvard was required by state laws to have insurance for workers’ compensation claims. To satisfy this express requirement, Harvard purchased insurance and paid premiums in the 1996 tax year. The fact that Harvard bought retrospective policies instead of traditional policies did not alter the nature of the payments.12
Harvard was charged additional premium expenses even after the policy expired. Those additional premiums were calculated using actual losses multiplied by a loss conversion factor. The loss conversion factor was designed in part to cover Wausau’s administrative costs, such as the cost of investigating and settling claims. The district court concluded that such features were always part of the price a taxpayer pays for insurance of any kind and thus should not be disqualified from the total qualifying specified liability loss because of the way it is calculated in this particular type of plan. We agree.
No insurance company would survive for long without covering its administrative costs. Those costs allow it to process and pay the claims that are the very purpose of purchasing an insurance policy. We see no justification in law or policy to allow these deductions for the actuarially derived cost of premiums and disallow the administrative costs attendant to every insurance policy merely because those costs are assessed and billed as they were here. We will therefore affirm the ruling of the district court in full on this issue.
V. CONCLUSION
For the reasons explained above, we reverse the district court’s ruling with respect to Harvard’s 1996 payments to its pension plans. On all other points, we affirm. We will remand this matter to the district court to recalculate the amount of refund due to the Trust in accordance with this opinion.