Deutsche Bank Ignored Some "Losses" Until They Went Away

Publish date:

Oh man, what is going on in this FT article? Here is the bottom line:

In a series of complaints to US regulators, two risk managers and one trader have told officials that Deutsche had in effect hidden billions of dollars of losses.

“By doing so, the bank was able to maintain its carefully crafted image that it was weathering the crisis better than its competitors, many of which required government bailouts and experienced significant deterioration in their stock prices,” says Jordan Thomas, a former US Securities and Exchange Commission enforcement lawyer, who represents Eric Ben-Artzi, one of the complainants.

The "in effect" does a lot of work there; Deutsche Bank "in effect" hid billions of dollars of losses because there were no losses. Other than that!

Here's a synopsis of what seems to have been going on:

  • Starting in 2005, Deutsche did some credit trades where they bought protection from some Canadian pension funds and sold protection to hedge funds, etc.
  • The bought and sold protection were not identical, with various technical bits of non-overlap that you can read about at your leisure down below.1
  • A credit crisis occurred, changing the risks involved in those non-overlapping bits from silly, abstract, purely theoretical risks into significantly more alarming and more-likely-to-occur but still purely theoretical risks.2
  • Deutsche's people sort of ran around dopily trying to figure out what to do about it. Here's a condensed version of the running around they did about the main risk, the "gap option" that DB was short in its leveraged super senior trades:

[T]he bank had at one point used a model but found it came up with “economically unfeasible” outcomes. Instead, it used two other measures. First, a 15 per cent “haircut” on the value of the trades … In 2008, during the crisis, instead of increasing the haircut, the bank scrapped it. The gap risk was now supposed to be covered by a reserve. The complainants say that the total of reserves held by the credit correlation desk was just $1bn-$2bn, which was supposed to cover all risks, not just the gap option. … Then, in October 2008, … Deutsche stopped any attempt to model, haircut or reserve for the gap option but says that the company took that action because of market disruption during the financial crisis. … At this time, to account for the gap risk, the bank hedged it by buying S&P “put” options.

  • They got through things just fine since after all they'd basically bought protection against world-ending correlated defaults in investment-grade companies and those defaults didn't happen:

“The valuations and financial reporting were proper, as demonstrated by our subsequent orderly sale of these positions,” Deutsche says. … A person familiar with the matter says that for all the sturm und drang over gap risk, at no time was the collateral jeopardised.3

  • No harm, no foul, but still a little scary, so after things settled down they did what they could to beef up their risk management for the next crisis, including hiring Eric Ben-Artzi, a math Ph.D. who'd worked at Goldman, in 2010.
  • Among his other talents, Ben-Artzi had at Goldman used models for some of these risks that were (1) more justified by the theoretical literature and (2) more conservative.
  • Then he went around being a dick about how they'd handled their risks ages ago - "For several months, Mr Ben-Artzi quizzed colleagues at Deutsche on how it modelled the gap option."
  • In 2006-2008 mind you. That is, for several months in 2010, this guy was quizzing people about the quality of their work during the financial crisis several years ago, when he wasn't there.
  • You'd have fired him too.
  • They did.
  • He went to the SEC, as did two other former DB employees, alleging that the various kerfuffling on the Deutsche correlation desk amounted to accounting fraud.

So: did it? I don't know. Is that an interesting question? Not really? A bank is just a collection of contracts entitling it to future cash flows in various states of the world; you don't really know what those contracts are worth until those states of the world obtain, so you don't really know what a bank is worth, ever. I submit that that - not "ooh they mismarked the gap option by $12 billion vs. the preferred model in the literature" or whatever - has to be your starting point in thinking about bank accounting.

When you start there, your accounting is pretty much "whatever the rules say and your auditors allow." Some things get marked to an objective market even if that makes you sad - the publicly traded stocks in your cash equities trading book, for instance. Some things get marked to historical cost, with some fudge-y reserve, even if that looks crazy - your whole-loan mortgages, for instance. There are good arguments that banks should mark more things to market, and there are good arguments that they should mark fewer things to market.

But some things live in a level-3 world where there are no specific rules and no obvious markets and you rely on good faith and estimates and conversations with your auditors. Tailored credit derivatives where Deutsche Bank was "65 per cent of all leveraged super senior trades" clearly fall in that bucket, so they talked to their auditors and the auditors signed off on what they did and, that's kind of the end. Was the model (or series of ad-hoc models and reserves and absences-of-models) blessed by the auditors worse than Ben-Artzi's? Sure! But so what? The model blessed by the auditors for loans - "hold them at historical cost" - is clearly worse than a market-based model, in the sense of "less reflective of the expected probability distribution of future cash flows." And loans are the bulk of most banks' assets. Accounting isn't supposed to be a correct representation of your most likely future cash flows. It's just accounting.

But this isn't just a story about accounting. Here's the FT on Ben-Artzi and the gap option:

Coming from Goldman, where the bank had modelled this risk, Mr Ben-Artzi believed this element of the trade was not inconsequential at all. Based on the model used at Goldman, he calculated it could have been worth as much as 8 per cent of the notional value of the trade during the crisis, or $10.4bn. When credit spreads deteriorate, he knew, banks should not just book the mark-to-market profit from the increased value of their protection but also the gap option: the mark-to-market losses associated with the counterparty walking away.

Now he's a "whistleblower" so this is all expressed in terms of booking and accounting but it isn't fundamentally that. It's really a question of trade selection. If you just ignore this way-out-of-the-money gap option, then this trade is free money,4 and you will do it all day long. If you value it using customary theoretical principles, you will do it for a much smaller portion of the day. (If you value it using Nassim Taleb principles,5 you will take the other side of it.) There's a reason Deutsche Bank was 65% of this market, and Goldman was … not.

And, y'know, good for Goldman, bad for Deutsche.6 Banks are supposed to take only economically justified risks, rather than just picking up proverbial pennies in front of proverbial steamrollers. Also: good for Deutsche for hiring Ben-Artzi to fix it, and bad for them for firing him. And probably bad for him for spending his time bashing around quizzing people about the past instead of preparing for the next crisis. I guess he's not alone in that: regulatory, journalistic, and whistleblower-compensation incentives all point towards making this a story about "accounting fraud in the past," rather than a story about "how banks should manage their risks in the future." The crisis-era-accounting story seems to me like a much less interesting, and less useful, story.

Deutsche Bank: Show of strength or a fiction? [FT]
Deutsche hid up to $12bn losses, say staff [FT]
The Deutsche allegations [Reuters / Felix Salmon]
Update:The ghost of the leveraged super senior [FTAV / Lisa Pollack]

1.The ones named in the FT article:

  • The Canadian pension funds didn't have to fully collateralize their trades; they were ~10x levered. This means that on a gap up in credit spreads, Deutsche would still be on the hook on its sold protection, but the Canadians might walk away from Deutsche's now-undercollateralized bought protection. The article refers to this as the "gap option" because in effect the pension funds were long an option to walk away: the contract gave them risk of loss from 30 to 100 (or whatever), but since they only had to post 10 of collateral the CSA in effect gave them risk of loss from 30 to 40 (or whatever). That's a put option and you can value it using models similar to the ones you use to value the tranche in the first place.
  • There was name mismatch: "Deutsche was buying protection on a customised range of companies, which included diverse names from around the world. It was then selling protection on a range of companies in the CDX index of US-only companies." Maybe they'd be correlated, maybe not.
  • There was a quanto risk: DB bought protection in Canadian dollars and sold it in US dollars, meaning that if spreads widened as USD strengthened vs. CAD, DB would owe more USD on its sold protection while getting fewer USD in from its Canadian counterparties. There is something to be said for this:

2.The simple way of supporting "still purely theoretical" is that the bought protection was all leveraged super senior tranches and so DB didn't have to care about collateral mismatches, etc. unless there were actually a lot of defaults on the names it had bought protection on. There's no list of those names - these were custom tranches apparently - but, come on. A quick look at CDX NA IG 5 on Bloomberg (CDX IG CDSI S5 10Y Corp MEMC) shows only 11 companies (CIT Group, Constellation Energy, Countrywide, Duke Energy, Freddie Mac, Fannie Mae, IAC, Knight Ridder, Radioshack, WaMu, XL Group) whose CDS are no longer trading; most of these are mergers rather than, y'know, WaMus. Even if they were all credit events, though, that's under 10% of an equally-weighted index; on a 30-100 tranche that has no effect. I mean it has a mark-to-market effect obvs, but not a final payout effect.

3.WAIT: is "sturm und drang" an FT interpolation, or do people at Deutsche Bank really talk like that? Please let them talk like that. Also: if you're German, is "sturm und drang" just a normal everyday thing to say? Like, "ugh, my commute was a nightmare, there was all sorts of sturm und drang on the Q this morning."

4.The option has a value equal to $X. A Canadian pension fund will pay you $X for it. You will book its value at $0. You have made $X of free money. QED.

5.I feel like that link needs a NSFW warning.

6.You could I suppose ask the contrarian question: was DB's model worse than Ben-Artzi's and Goldman's? I mean, setting aside that DB didn't seem to have a model so much as a series of uncoordinated guesses. DB modeled the chances of blowing through its collateral on these trades at, roughly, zero; and it didn't blow through its collateral on these trades. Was its probability wrong? Oh, sure, I mean, go read this paper about estimating that probability; I won't stop you. But at the end of the day it kind of looks like (1) Deutsche Bank thought it could buy credit protection a little cheaper by including a gap option in the contract, (2) it thought that gap option would not turn out to have any value at expiry, and (3) it was right on both counts. So Goldman, Ben-Artzi et al. shouldn't be too smug. The job of a bank is not to impress quantitative-finance researchers.

Also, DB isn't alone. Here's a passage from the FT about another of the risks here:

Eventually, Deutsche reached for a saviour that had helped many institutions during the financial crisis: Mr Buffett. … Berkshire wrote insurance on the quanto risk for Deutsche at a cost of $75m in 2009. Deutsche then accounted for this as full protection on the risk. But the contract agreeing the trade, reviewed by the FT, caps the payout in the event of losses at $3bn, while Deutsche was claiming protection on tens of billions of dollars. Once again, the former employees allege, the bank was accounting as if it had fully insured itself against loss while in reality insuring itself against only a small portion.

So, yeah: what dolts DB were, totally ignoring the possibility that the quanto risk would cost them more than $3 billion! But here's how I'd do the math on that probability. If the losses exceed $3bn, then Berkshire is on the hook for $3bn. It got paid $75mm. Think what you will about Warren Buffett's sagacity as a trader of derivatives,a but: How likely is it that Berkshire would agree to a contract that had a high probability of requiring it to pay out 40x what it got paid in premiums?b I submit to you that if Deutsche bought first-$3bn protection from Berkshire for $75mm, that first $3bn of protection was worth less than $75mm, which makes the next, like, $100bn of protection worth really not that much at all.c

a.Also about the likelihood that he was personally involved in negotiating this trade.

b.I know, I know, home insurance or whatever, but there's not really a portfolio effect here.

c.Ooh, don't put too much stock in that last bit of pseudo-math. Correlation! Fat tails! Etc. Still.