Banque Worms
Last August, Citigroup Inc. wired $900 million to some hedge funds by accident. Then it sent a note to the hedge funds saying, oops, sorry about that, please send us the money back. Some did. Others preferred to keep the money. Citi sued them. Yesterday Citi lost, and they got to keep the money. I read the opinion, by U.S. District Judge Jesse Furman, expecting to learn about the New York legal doctrine of finders keepers—more technically, the “discharge-for-value defense”—and I was not disappointed. But I was also treated to a gothic horror story about software design. I had nightmares all night about checking the wrong boxes on the computer.
The story—we have discussed it before—is that, in 2016, Revlon Inc. took out a seven-year syndicated term loan. Citibank N.A. is the administrative agent on the loan; it gets interest and principal payments from Revlon and passes them on to the lenders. Revlon ran into a bit of trouble and, as companies do these days, it did some creative stuff with its debt: In May 2020, it convinced some of the term-loan lenders to strip collateral from the term loan so it could be used to back new debt. The lenders who were part of this “incredibly aggressive” deal got to roll over into the new, effectively more senior debt; the other lenders were left with worse debt and got mad. Some of them got together to work on a lawsuit, which they filed on Aug. 12.
Twenty hours before they filed the lawsuit, though, they got lucky: Citigroup just wired them all their money. They received wire transfers for the full amount of principal and accrued interest they were owed on the loan. Their first reaction was mostly “well this is weird, I guess Revlon decided to pay off the loan rather than fight about it.” Their second reaction, after Citi sent them frantic notices saying it was a mistake, was to send each other Bloomberg chat messages making fun of Citi. Their third reaction, after some more serious reflection, was to say “we are keeping the money, see you in court.” All of these reactions were pretty reasonable and worked out well for them.
What happened? Well, it starts with the fact that some of the term-loan lenders had agreed to the aggressive deal to put in new money and roll their term loans into new, better-secured debt. 1 So they came to Citi and Revlon, handed in their old debt and got back new debt. When they do this, customarily, they get paid accrued interest on their old debt. Citi, for some reason, couldn’t handle that sensibly; from the opinion:
Given certain technical limitations of Citibank’s system for making payments, the most efficient way for Citibank to effect the transaction was to pay interim interest accrued to all lenders that held 2020 Extended Term Loans; paying only the rolling-up entities would have required a “very manual process.”
So instead of just paying interim interest to the lenders who were rolling their old loans into new loans, Citi had to pay it to all of the lenders, and Revlon agreed to make an interim interest payment to everyone. 2 So Revlon wired $7.8 million—for an interest payment—to Citi, and Citi got set up to pay it to the lenders: 3
The August 11th roll-up transaction involved five Lenders, all managed by Angelo, Gordon and Co. (“Angelo Gordon”). The Lenders affiliated with Angelo Gordon were exchanging their positions in the 2016 Term Loan for positions in a different Revlon credit facility. Following this exchange, the remaining Lenders would continue to hold a pro rata share of the 2016 Term Loan on a slightly reduced principal balance. As noted above, when a lender rolls up and exchanges a position in one credit facility for another, it is typically paid the accrued interest on the first facility at the time of the exchange. Due to the same technical limitations of Citibank’s system ... Revlon agreed to pay accrued interest to all 2016 Term Loan Lenders to effect the Angelo Gordon roll-up transaction — even though the other Lenders were not involved in the roll-up transaction and even though an interim interest payment was not due under the Amended Loan Agreement until August 28, 2020.
But the Angelo Gordon funds were getting taken out of the loan entirely and rolled into the new facility, so their principal also had to be paid off. (Not really—they would get cashed out at par and roll their money into the new facility, without taking out actual cash—but as a bookkeeping matter.) Here is a paragraph that I think you can only read with slowly dawning horror:
Citibank’s Asset-Based Transitional Finance (“ABTF”) team, a subgroup of Citibank’s Loan Operations group that is focused on processing and servicing of asset-based loans, was tasked with executing the roll-up transaction on Flexcube, a software application and loan product processing program that the bank uses for initiating and executing wire payments. On Flexcube, the easiest (or perhaps only) way to execute the transaction — to pay the Angelo Gordon Lenders their share of the principal and interim interest owed as of August 11, 2020, and then to reconstitute the 2016 Term Loan with the remaining Lenders — was to enter it in the system as if paying off the loan in its entirety, thereby triggering accrued interest payments to all Lenders, but to direct the principal portion of the payment to a “wash account” — “an internal Citibank account that shows journal entries . . . used for certain Flexcube transactions to account for internal cashless fund entries and . . . to help ensure that money does not leave the bank.”
Ah ha ha! Yes! The “easiest (or perhaps only)” way to pay off some lenders but not others was to instruct the software to pay off all the lenders! But tell it only to pretend to pay them! Just send that money to a wash account! This is all fine! Let’s read another horrifying paragraph!
Because the vast majority of wire transactions processed by Citibank using Flexcube involve the payment of funds to third parties, any payment entered into the system is released as a wire payment unless the maker suppresses the default option. Citibank’s internal Fund Sighting Manual provides instructions for suppressing Flexcube’s default. When entering a payment, the employee is presented with a menu with several “boxes” that can be “checked” along with an associated field in which an account number can be input. The Fund Sighting Manual explains that, in order to suppress payment of a principal amount, “ALL of the below field(s) must be set to the wash account: FRONT[;] FUND[; and] PRINCIPAL” — meaning that the employee had to check all three of those boxes and input the wash account number into the relevant fields.
This is just demented stuff. If you want to send out interest payments in cash, but send the principal payment to the wash account, you have to check the box next to “PRINCIPAL” and also the boxes next to “FRONT” and “FUND.” “PRINCIPAL” sounds like principal: You are sending the principal to the wash account, sure, right, yes, check that box. “FRONT” and “FUND” sound like nothing. So the Citi operations people messed it up:
Notwithstanding these instructions, Ravi, Raj, and Fratta all believed — incorrectly — that the principal could be properly suppressed solely by setting the “PRINCIPAL” field to the wash account. Accordingly, as Ravi built out the transaction between 5:15 and 5:45 p.m. in his role as maker, he checked off only the PRINCIPAL field, neglecting the FRONT and FUND fields. Figure 1, below, “is an accurate image of the Flexcube screen after [Ravi] input the data.”
At 5:45 p.m., Ravi emailed Raj for approval of the transaction, explaining that “Princip[al] to Wash A[ccount] & Interest to DDA A[ccount].” The “DDA Account” referenced the Demand Deposit Account, which is an operational, external-facing account used by Citibank to collect payments from customers and make transfers to lenders. After reviewing the transaction, Raj believed — incorrectly — that the principal would be sent to the wash account and only the interest payments would be sent out to the Lenders. Raj then emailed Fratta, seeking final approval under the six-eye review process, explaining “NOTE: Principal set to Wash and Interest Notice released to Investors.” Fratta, also believing incorrectly that the default instructions were being properly overridden and the principal payment would be directed to the wash account, not to the Lenders, responded to Raj via email, noting, “Looks good, please proceed. Principal is going to wash.”
The software gave him a warning, but not a very good one:
Raj then proceeded with the final steps to approve the transfers, which prompted a warning on his computer screen — referred to as a “stop sign” — stating: “Account used is Wire Account and Funds will be sent out of the bank. Do you want to continue?” But “[t]he ‘stop sign’ did not indicate the amount that would be ‘sent out of the bank,’ or whether it constituted an amount equal to the intended interest payment, an amount equal to the outstanding principal on the loan, or a total of both.” Because Raj intended to release “the interim interest payment to [the] [L]enders,” he therefore clicked “YES.”
Here’s Figure 1; it does not particularly explain itself:
See, the “don’t actually send the money” box next to “PRINCIPAL” is checked, but that doesn’t do anything, you have to check two other boxes to make it not actually send the money.
When they discovered the error the next day, their first reaction was not to email the lenders asking for the money back (that was their second reaction); their first reaction was to email tech support to say the software was broken:
At 10:26 a.m., Fratta emailed Citibank’s technology support group: “Yesterday we processed a payment with Principal to the wash and Interest to be sent to lenders. All details in the front end screens yesterday le[d] us to believe that the payment would be handled in that manner. . . . Screenshots provided below indicating that the wash account . . . is present and boxes checked appropriately for the principal components.” Fratta then forwarded the same email to members of his team, with the subject line “Urgent Wash Account Does not Work.” He stated: “Flexcube is not working properly, and it will send your payments out the door to lenders/borrowers. The wash account selection is not working. This lead [sic] to ~1BN going out the door in error yesterday for an ABTF Deal, Revlon.” ...
Over the course of the day, Fratta learned that the principal payments — which were made with Citibank’s own money, as Revlon had provided funds only for the interim interest payments to be made in connection with the roll up transaction —were not caused by a technical error, but by human error: the failure to select the FRONT and FUND fields when inputting the default override instructions in Flexcube.
Nope, nope, he was right the first time, this whole setup is a “technical error.” Citi’s software will only let you pay principal to some lenders if you pretend to pay it to every lender, and it will only let you pretend to pay principal to every lender if you check the “just pretend” box next to “PRINCIPAL” (fine!) and “FUND” (what?) and “FRONT” (what even?). What a terrifying thing.
Anyway so, right, clearly it was a mistake, and Citi asked for its money back. It wired about $900 million of mistaken principal payments, and funds that got about $500 million refused to give the money back. “Finders keepers” is not actually a rule of New York law, and in general if you get a mistaken wire transfer you have to give it back. Citi sued, and the funds said, well, we were owed this money, and you sent it to us, so we’re going to keep it. The legal doctrine—the exception to the general rule that you have to give back mistaken wire transfers—is called the “discharge-for-value defense”:
The recipient is allowed to keep the funds if they discharge a valid debt, the recipient made no misrepresentations to induce the payment, and the recipient did not have notice of the mistake. As the New York Court of Appeals explained the exception: “When a beneficiary receives money to which it is entitled and has no knowledge that the money was erroneously wired, the beneficiary should not have to wonder whether it may retain the funds; rather, such a beneficiary should be able to consider the transfer of funds as a final and complete transaction, not subject to revocation.”
The leading case is called Banque Worms, which sounds right. When a bank wires someone money by mistake it can say “ugh we’ve got the banque worms again.”
Honestly it is a very strange doctrine. Here it makes some rough sense because the lenders had a real argument that Revlon had defaulted on the loan (by doing the aggressive collateral-stripping transaction), so it was immediately due and payable, but that’s not actually a requirement of the discharge-for-value defense and isn’t really discussed in the opinion. 4 If everything was fine with the Revlon loan, the lenders had no complaints, and Citi accidentally wired them the money, they’d still get to keep it. 5
Much of the dispute in this case is about whether “the recipient(s) did not have notice of the mistake,” that is, whether the lenders should have known, or did know, that the wire transfers were a mistake when they got them. They argued that they had no idea anything was wrong, that the payments were the exact amounts they were owed, that they assumed Revlon was intentionally paying down its loan to avoid litigation or do some other weird transaction, and that it didn’t cross their mind that Citi had messed up until Citi sent them recall notices the next afternoon.
Once Citi did send the recall notices, of course, the lenders knew it was a mistake, and they all sent each other chat messages making fun of Citi. “Not surprisingly,” writes Judge Furman, “given the nature and size of the mistake, many of these were quite colorful.” He can’t resist quoting some funny ones and neither can I:
DFREY5: I feel really bad for the person that fat fingered a $900mm erroneous payment. Not a great career move
. . . .
JRABINOWIT12: certainly looks like they’ll be looking for new people for their Ops group
DFREY5: How was work today honey? It was ok, except I accidentally sent $900mm out to people who weren’t supposed to have it
DFREY5: Downside of work from home. maybe the dog hit the keyboard
JRABINOWIT12: the song “Had a Bad Day” playing the background
But the judge points out that these chats only happened after the recall notices went out, and “the number and nature of these communications reinforce why the absence of such communications before the Recall Notices is so significant.” That is, if the lenders had thought the payments were a mistake when they got them, they would have been unable to resist hopping into a chat room and cracking jokes about Citi, as proven by the fact that when they got the recall notices they did all crack jokes about Citi. The fact that they didn’t make any jokes for almost a full day proves that, when they got the payments, they thought they were legit.
It is a weird rule that, if you get a payment that you think is legit, and then one minute later you get a notice saying “no sorry this payment was an error,” you nonetheless get to keep the payment, but I guess that’s the rule. Banque Worms!
What a mess. Obviously this is good for the funds who kept the money. It is awkward for the funds who returned the money; they can’t ask Citi to send it back to them. They’re stuck holding the loan until it matures or defaults; Bloomberg tells me that it’s trading at around 42 cents on the dollar. Their clients are going to have questions about their aggressiveness and creativity. Aggressiveness and creativity are kind of the whole ballgame when you are trading distressed debt; the business is about hunting for arcane advantages that you can exploit to get more money than the other guys. In a sense the discharge-for-value exception is an arcane advantage, but in another sense “well they sent us money so we’re going to keep it” is the least arcane imaginable thing, and if you don’t have that instinct perhaps you were meant for a gentler corner of the financial world.
It’s awkward for Citigroup and Revlon too. What do they do? Does Revlon owe Citi the $500 million now? Payable in 2023? I mean, presumably, right? 6 Presumably when Citi accidentally paid off Revlon’s loan, that wasn’t just a gift to Revlon? But neither did it accelerate Revlon’s debt? Citi just bought $500 million worth of the term loan at par and has to wait to get paid back? “‘If appeals fail, Citi will ultimately step into the shoes of the lenders and own $500 million of that nearly $900 million term loan,’ said Philip Brendel, a senior distressed debt analyst at Bloomberg Intelligence.” Can it syndicate whatever this is? Sell some of its weird phantom claims on the term loan to distressed-debt funds? Maybe the same funds that just took it for $500 million? Or I suppose Citi and Revlon could cut a deal where Citi gets paid back X cents on the dollar soon and leads some weird new debt-restructuring transaction for Revlon to fund it. That should be easier now. All the aggressive funds are gone.
When this all happened, there was a certain amount of commentary to the effect of This Proves Citi Is Too Big To Manage and a Threat to Global Financial Stability. That feels a little overblown to me—meh, now Citi owns $500 million of a mispriced loan to Revlon, it’s had bigger problems—but on the other hand, what an absolutely hair-raising description of Citi’s software this is. It is all well and good to say that a bank is “too big to manage,” but what that means in practice is surely something like this. It means you have to check three boxes to not send out money instead of one, and people forget to check two of them.
https://www.bloomberg.com/opinion/articles/2021-02-17/citi-can-t-have-its-900-million-back....
Know how some critics complain bitcoin accidentally sent to the wrong address is non recoverable?
Citigroup *accidentally* sent $900 million to hedge funds. And was unable to recover $500 million of their accidentally sent transaction.
The case went to court and the judge ruled against citigroup. So hedge funds were allowed to keep the free money.
This illustrates how even banks who send transactions to the wrong recipients aren't necessarily better off than someone making identical errors in crypto.
