LOKEN, Circuit Judge.

John Markert, while President of Pinehurst Bank ("the Bank" or "Pinehurst"), approved five nominee loans by the Bank to friends and family of bank customer George Wintz. The loan proceeds were used to cover a nearly $1.9 million overdraft in Wintz's checking account at the Bank. A jury convicted Markert of willful misapplication of bank funds by a bank officer in violation of 18 U.S.C. § 656. At sentencing, applying Application Note 3 to U.S.S.G. § 2B1.1(b)(1), the district court found that Markert's offense caused an actual loss to the Bank equal to the total amount of the nominee loans, resulting in a 16-level enhancement and an advisory guidelines range of 87 to 108 months in prison. The court granted a substantial downward variance and sentenced Markert to 42 months in prison. Markert appealed his conviction and sentence. A divided panel affirmed the conviction, unanimously concluded that actual loss was erroneously calculated, and remanded for resentencing. United States v. Markert, 732 F.3d 920 (8th Cir.2013).

On remand, the government again asserted that actual loss for purposes of § 2B1.1(b)(1) was the face amount of the nominee loans, subject only to a credit for $60,000 in principal repayments made prior to detection of the fraud, see § 2B1.1, comment. n. 3(E)(i). After considering arguments of counsel, but without an evidentiary hearing, the district court agreed with the government and reduced its prior finding of loss only by the amount of those repayments. This small reduction did not affect the 16-level enhancement, see § 2B1.1(b)(1)(I), and the court "re-impose[d] the same 42-month term of imprisonment previously ordered."

Markert again appeals that sentence. We again review the district court's fact findings for clear error and its interpretation of the advisory guidelines de novo. See United States v. Holthaus, 486 F.3d 451, 454 (8th Cir.), cert. denied, 552 U.S. 939, 128 S.Ct. 343, 169 L.Ed.2d 241 (2007). As explained below, the government misinterpreted (or refused to follow) the relevant loss principles discussed in our prior opinion. Markert, 732 F.3d at 931-33. As a result, the government failed to sustain its burden to prove actual loss, and we must again remand for resentencing. Though we agree with the district court that "the loss here cannot be zero," we decline to give the government a third chance to present evidence meeting its burden of proof. Thus, on remand, actual loss for sentencing purposes is zero, reducing the advisory guidelines range to 12-18 months in prison. As Markert has already served more than 18 months, we reverse the district court's Order dated December 26, 2013, remand with instructions to sentence Markert to time served, and direct that he be immediately released from prison, subject to the terms and conditions of supervised release imposed by the district court.

I. Background

We summarize the evidence at trial relevant to the actual loss issue. Additional background may be found in our prior opinion. After Markert became Pinehurst's President in 2007, Wintz opened business checking accounts for two of his trucking and warehouse entities, McCallum Transfer and Cue Properties. Markert approved a series of loans to Wintz, quickly reaching the $250,000 limit of Markert's unilateral lending authority. By February 2009, loans to Wintz had reached the Bank's legal lending limit of nearly $1.2 million. JoAnn Crowley, Pinehurst's Chief Financial Officer, repeatedly warned Markert that Wintz may be "kiting checks."1 Her warnings went unheeded.

In early March, when employees at Northstar Bank discovered Wintz's check-kiting activities, Northstar dishonored fifteen checks totaling nearly $1.9 million drawn on a Wintz account at Northstar and deposited into his McCallum Transfer account at Pinehurst. With the deposit checks dishonored, Pinehurst faced a nearly $1.9 million loss on Wintz's overdrawn account. By Monday, March 9, Wintz had persuaded five friends and family members to sign documents obligating them to repay five loans by Pinehurst totaling $1.9 million. Each nominal borrower understood that Wintz was the real borrower and would be responsible for principal and interest payments. Markert and others at the Bank prepared and closed the five nominee loans on Monday, March 9. Disguised as investments in Cue Properties, the loan proceeds were credited to Wintz's Cue Properties account, then immediately re-directed and credited to his McCallum Transfer account. The Bank did not post the checks returned by Northstar until March 10 and did not record the $1.9 million overdraft because by then the loan proceeds had infused Wintz's account with sufficient funds. Markert did not tell the Bank's Board of Directors that Wintz was the real borrower on the five nominee loans, nor disclose Wintz's near-overdraft and check-kiting activities.

In January 2010, during a routine audit, an auditor uncovered the true purpose of the five nominee loans. Markert was immediately terminated. In February 2010, after reviewing the nominee loans, bank regulators required the Bank to book an additional $2.2 million in loan reserves. In early April, the Bank as lender, Cue Properties as borrower, and Wintz as guarantor entered into a fully collateralized Loan Consolidation and Modification Agreement releasing the nominee borrowers from their obligations. That agreement made the Bank whole from Markert's misapplication offense, but the FDIC's continuing investigation uncovered additional loss exposures, and regulators closed the Bank in May 2010.

Markert was convicted of violating 18 U.S.C. § 656, an offense requiring proof that he "wil[l]fully misapplied funds for the benefit of himself or another person, for the purpose of defrauding or injuring" Pinehurst Bank. United States v. Barket, 530 F.2d 181, 186 (8th Cir.1975), cert. denied, 429 U.S. 917, 97 S.Ct. 308, 50 L.Ed.2d 282 (1976); see United States v. Britton, 107 U.S. 655, 666-67, 2 S.Ct. 512, 27 L.Ed. 520 (1883). The nominee loan proceeds were funds of the Bank. Here, as in other cases where a bank officer used nominee loans to camouflage fraudulent transactions, the "other person" — check-kiter Wintz — was the real borrower, not the nominee borrowers.2 Wintz gained control of the nominee loan proceeds deposited into his account and benefitted by avoiding a large overdraft that would have severely damaged his business and financial interests. His contractual relations with the Bank gave the Bank immediate right to use the loan proceeds to cover the deposit checks that had been dishonored by Northstar Bank.

II. Discussion

The Guidelines define "loss" for purposes of the enhancement for fraud offenses as "the greater of actual loss or intended loss." U.S.S.G. § 2B1.1, comment. n. 3(A). Under this guideline as extensively amended in 2001, "actual loss" is "the reasonably foreseeable pecuniary harm that resulted from the offense." Note 3(A)(i); see United States v. Hartstein, 500 F.3d 790, 798 n. 3 (8th Cir.2007), cert. denied, 552 U.S. 1102, 128 S.Ct. 939, 169 L.Ed.2d 734 (2008). As the Sentencing Commission explained, this "net loss" approach "recognizes that the offender who transfers something of value to the victim[ ] generally is committing a less serious offense than an offender who does not." U.S.S.G.App. C., Vol. II, Amend. 617, at 183; accord § 2B1.1 comment. (background). Pinehurst Bank was the only victim of Markert's offense. The government concedes he did not intend to cause Pinehurst a loss. Thus, the issue is actual loss, which government has the burden to prove by the preponderance of the evidence. "The court need only make a reasonable estimate of the loss." § 2B1.1, comment. n. 3(C).

Markert's Revised Presentence Investigation Report recommended, without explanation, that Markert's willful misapplication offense caused an actual loss to the Bank equal to the total amount of the five nominee loans, nearly $1.9 million. The government's sentencing memorandum supported this recommendation, arguing (i) numerous cases have held that loss in check-kiting cases is calculated in this fashion, and (ii) "the amount of loss in misapplication and fraudulent loan cases is the amount of funds misapplied." The district court adopted the government's argument, explaining: "The amount of funds misapplied is the amount of the loss in a misapplication case and here those nominee loans totaled $1.8 million." On appeal, we concluded that "the government's actual loss contention as adopted by the district court ... fail[ed] to acknowledge that the monetary value of the nominee loans the Bank received in exchange for the misapplied proceeds, measured at the time the misapplication offense was detected, must be credited against actual loss." Therefore, we remanded to determine "[t]he net value of those loans, measured at the time their nominal nature was detected." 732 F.3d at 932-33. Judge Bright in dissent noted his view that "the true degree of actual loss in this case [is] zero." Id. at 935.

On remand, the government succinctly stated its position at the resentencing: "From our perspective, when Mr. Markert learned of this check-kiting scheme and determined to secretly deploy the bank's capital to fund that scheme, that constituted an actual loss to the bank of $1.9 million at the time the funds were disbursed." This was the same position the government took at the initial sentencing. We rejected this method of determining the actual loss caused by this willful misapplication offense, noting that the government supported its contention with only one prior case that we found distinguishable "because the misapplied funds at issue were not the proceeds of new loans by the victim bank." Id. at 932 (emphasis in the original). The government inexplicably ignored this directive to rethink its initial position.

For purposes of the § 2B1.1(b)(1) enhancement, actual loss is "reasonably foreseeable pecuniary harm," that is, "harm that is monetary or that otherwise is readily measurable in money." Note 3(A)(i) & (iii). For many, perhaps most fraud offenses, actual loss is properly and readily measured by the fair market value of property "taken" from the victim. Note 3(C)(i). That will be true in some willful misapplication cases, such as United States v. Thomas, 422 F.3d 665, 667 (8th Cir. 2005), where the misapplication offense was causing the bank to disburse loan proceeds to a borrower who defaulted. But in this case, the money never left the Bank. In affirming Markert's conviction, we reviewed many cases establishing that "it is not always necessary that money should be actually withdrawn from a bank, to constitute a criminal misapplication of its funds." 732 F.3d at 927 (quotation omitted). But that does not resolve the question of actual loss for sentencing purposes. Cf. United States v. Sandison, No. 09-CR-0108, 2011 WL 1596205 at *8 (D.Minn. Apr. 22, 2011). When a willful misapplication offense does not result in money being "taken" from the bank, as here, actual loss needs to be determined another way, for example, by estimating "[t]he reduction that resulted from the offense in the value of ... corporate assets." Note 3(C)(v). It may not be estimated simply by declaring, without supporting evidence, that actual loss was the total amount of the assets that were willfully misapplied from one segment of the bank's balance sheet to another (here, the nominee loans moved dollars from capital available for lending to an overdrawn customer's checking account deposits).

We have little doubt the Bank suffered some actual pecuniary loss as the result of Markert's willful misapplication of the nominee loan proceeds. There are many ways that pecuniary harm could be proved. The most obvious is, as our prior opinion explained, to determine "[t]he net value of those [nominee] loans, measured at the time their nominal nature was detected." 732 F.3d at 933. That value is not, as the government argues, a determination of the amount to credit against an actual initial loss of $1.8 million. As the loan proceeds were not initially "taken" from the Bank, actual loss was only the difference between the amount loaned and the value of the loan (to the extent not repaid) when the fraud was detected. As we observed, that amount "may be difficult to measure." Id. at 933. But the bank officers who succeeded Markert could have estimated the net value of the nominee loans at that time, perhaps with help from an expert in valuing distressed bank loans.

Here, the record on appeal contains no such evidence. All the record tells us is that, three months after the fraud was detected, Wintz agreed to a loan "modification" agreement that provided the Bank a fully collateralized loan, and released the nominee borrowers and their collateral from the Bank's dubious claims against them.3 The government urges us to ignore this fact because "a defendant's repayment of funds after the discovery of a fraud offense is not relevant to sentencing." Id. at 932-33 n. 5. That is generally true. But here, the Modification Agreement was the best evidence of the net value of the nominee loans three months earlier when the government failed to offer any other evidence.

The government could also have introduced other evidence showing pecuniary harm to the Bank. For example, the Bank doubtless incurred attorney's fees and other expense in convincing Wintz to enter into the Modification Agreement rather than face lawsuits by the Bank and by the nominee borrowers, and other harm to his business interests. In addition, the Bank suffered foreseeable regulatory harm from the willful misapplication offense; the government could have presented evidence that adverse regulatory consequences caused pecuniary harm.

By accepting the government's contention, the district court assumed that the nominee loans had zero net value on the date of detection.4 The government suggests this assumption was appropriate because the FDIC classified the nominee loans as "loss loans." We disagree. In the first place, the FDIC reclassification was stated in the PSR, but it was not proved, and its pecuniary significance was not explained, at trial. More importantly, that a regulatory agency classifies a bank loan as "loss" or "uncollectible" when fraud is detected hardly establishes that the loan has no value to the bank. Banks are accustomed to engaging in collection efforts to recover some value when they have made loans to financially stressed borrowers, or loans for which the borrower may have a defense. If a particular bank is unwilling to devote its resources to such efforts, there are businesses who will purchase doubtful loans at a discount and attempt to recover from the borrowers more than they paid the bank.

III.

To summarize, the government on remand ignored its burden to establish the reasonably foreseeable pecuniary harm resulting from Markert's offense. It simply declared that over $1.8 million was the actual loss and argued that the offsets or credits recognized in Application Note 3(E) did not apply. This put the cart before the horse. As the principal amount of the nominee loans was not a proper initial estimate of actual loss, whether to reduce the actual loss by the Credits Against Loss allowed in Application Note 3(E) became academic. In essence, the government's flawed argument to the district court impermissibly placed the burden on Markert to prove that actual loss was less than $1.8 million.5

The government has had two opportunities to prove the actual loss resulting from Markert's offense. Our prior remand explained what needed to be proved. While we doubt the actual pecuniary loss to the Bank was zero, the government failed to present evidence at trial or at sentencing permitting a reasonable estimate of that loss. We recognize this was a novel sentencing issue, but given the nature of the offense and the time Markert has served in prison, we conclude we should follow the "traditional path" and not give the government another "bite at the apple." United States v. Thomas, 630 F.3d 1055, 1057 (8th Cir.2011); see United States v. Gammage, 580 F.3d 777, 779-780 (8th Cir.2009). Therefore, no § 2B1.1(b)(1) enhancement may be imposed in any further sentencing proceeding. That reduces Markert's total offense level to 13 and his advisory guidelines range to 12-18 months in prison. He has twice received a substantial downward variance from the district court, so any sentence above this revised range would raise vindictive sentencing issues. Markert entered custody on October 31, 2012, and has served more than 18 months in prison. In these circumstances, we conclude the proper disposition is to remand to the district court with instructions to resentence Markert to time served, and we direct the appropriate authorities that he be immediately released.

Let the mandate issue forthwith.