Jérôme Kerviel and the €4.9 Billion Loss at Société Générale

On Saturday 19 January 2008 a small group of executives sat on an upper floor of Société Générale's towers at La Défense and tried to work out what one of their junior traders had actually bought. The previous afternoon a control officer had queried an unusually large counterparty exposure booked against a German bank. The German bank had no record of the trade, for the straightforward reason that the trade did not exist. By Sunday evening the reconstruction was complete, and it showed that Jérôme Kerviel, thirty-one years old and four years out of the middle office, was long roughly €50 billion of European equity index futures — a notional exposure larger than the stock market value of the bank that employed him.

Daniel Bouton, the chief executive, had two choices and no good ones. Disclosing the position before closing it would invite every trading desk in Europe to sell ahead of the unwind. Closing it quietly meant liquidating €50 billion of futures into a market that, on the Monday morning, was about to have its worst day since 2001. He chose the second. Over three sessions the bank sold out, and on Thursday 24 January it told the world it had lost €4.9 billion.

From the Middle Office to Delta One

Kerviel was not a product of the French financial elite. Born in 1977 at Pont-l'Abbé in Brittany, son of a hairdresser and a metalwork teacher, he took a masters in organisation and control of financial markets at Lyon rather than a place at one of the grandes écoles that supplied Société Générale's trading floor. He joined the bank in 2000 and spent five years in the middle office, in the compliance and control functions whose job is to check what traders do.

In 2005 he moved to the Delta One desk. Delta One trading covers instruments that track their underlying one-for-one — index futures, exchange-traded funds, swaps, certificates — and the desk's mandate was arbitrage: exploiting small price differences between a basket of shares and the future written on it, while carrying close to no directional exposure. The permitted net position was measured in tens of millions of euros. It was, in the bank's own description, plain-vanilla work, which was precisely why it attracted less scrutiny than the exotic derivatives desks two floors away.

His prior career was the problem. Kimberly Krawiec's study of rogue trading cases across three decades identifies the same structural feature in almost all of them: the trader either came from the control function, as Kerviel and Nick Leeson had, or had been left to supervise the recording of his own trades (Krawiec, 2009). Someone who has spent five years learning which reconciliation runs on which night, and how long a confirmation can stay outstanding before anybody asks, knows exactly where the gaps in the net are.

The Mechanics of a Hidden Book

What Kerviel did with his real position was simple. He took large directional bets that European equity indices would fall in 2007, and then that they would rise in early 2008. What he did with the record of that position was more careful.

Against each genuine futures position he booked a fictitious offsetting trade — a forward or an over-the-counter transaction with a counterparty chosen because trades of that type carried a settlement or confirmation lag. On the bank's screens the two legs netted to something close to zero, which is what the risk system expected a Delta One book to show. Before the fictitious leg aged far enough to trigger a confirmation request, he cancelled it and booked another. Provisional trades, technical cancellations, counterparties inside the group, trades dated to fall outside a reporting window: the toolkit was drawn from the operational routine he had spent five years administering. Investigators also found he had used colleagues' access codes and had produced forged email correspondence to satisfy queries.

Through 2007 the concealed position was profitable. By 31 December 2007 it carried an unrealised gain of roughly €1.4 billion, a sum Kerviel could not declare without revealing how it had been made, and which he masked in turn with a fictitious loss of similar size. His own pay that year came to something near €100,000. He had asked for a bonus of €600,000.

At the turn of the year he reversed the direction of the book, moving long into a market that was already falling. By mid-January the notional stood at about €50 billion, spread across Euro Stoxx 50, DAX and FTSE 100 futures, and the €1.4 billion gain had become a loss running into the billions.

Seventy-Five Alerts

The account of the affair eventually published by the bank's own committee of independent directors, working with PricewaterhouseCoopers and issued on 23 May 2008, is unusually candid about what the controls did see. Between June 2006 and early 2008 something on the order of seventy-five separate alerts, anomalies and queries were raised on Kerviel's activity (PwC, 2008). None was escalated to a level at which anyone asked the one question that would have ended it: not whether the trade was correctly booked, but whether the counterparty existed.

Eurex, the derivatives exchange where much of the position sat, had written to Société Générale in November 2007 asking about the size of its positions and the pattern of its trading. Kerviel helped draft the reply. A control function that checks the paperwork of a trade against the paperwork of its hedge, rather than against an outside party, is checking one man's arithmetic against his own.

Philippe Jorion's reading of the episode places it with Barings and Daiwa rather than with the model failures that were breaking other banks in the same month: nothing here required a mispriced instrument or a flawed value-at-risk assumption, only a nominal position that nobody measured on a gross basis (Jorion, 2008). Mark Wexler, writing on the persistence of rogue trading, makes a related point about incentives — a desk whose profits arrive without any corresponding use of balance sheet is a desk that has stopped being an arbitrage business, and the profit itself is the alert (Wexler, 2010).

Three Days in January

Jean-Pierre Mustier, who ran the corporate and investment bank, was given the unwind. The instruction was to sell without signalling: no more than a set share of volume in any contract, spread across three exchanges, with no explanation to anyone outside a handful of people at the top of the firm. The French market regulator and the Banque de France were told; the market was not.

Monday 21 January was a public holiday in the United States, which meant thin liquidity and no American buyers. European indices fell all day. The CAC 40 closed down 6.83 per cent at 4,744.45, its worst session since September 2001; the DAX lost 7.16 per cent. How much of that fall was Société Générale's selling and how much was the subprime panic arriving in Europe has never been settled, and the bank has always maintained its share of volume was modest. What is not in dispute is that the largest seller in European index futures that day was a bank liquidating a position it had not known it held.

Before the New York open on Tuesday 22 January the Federal Reserve cut the federal funds target by 75 basis points to 3.50 per cent, an emergency move between scheduled meetings of a size not used since the early 1980s. The committee was responding to the global rout of the previous day. Whether that rout had been amplified by an unwind in Paris that no central banker had been told about became, briefly, a question in Washington; Fed officials said afterwards that they had known nothing of it.

By the close on Wednesday 23 January the book was flat. The arithmetic the bank published the following day ran roughly as follows: a position showing €1.4 billion of unrealised profit at the year end, a liquidation loss of about €6.3 billion in January, and a net €4.9 billion written off against the 2007 accounts.

The Announcement

Thursday's statement carried three items at once. The trading loss was €4.9 billion. A further €2.05 billion was written down against subprime and monoline exposures. And the bank announced a €5.5 billion rights issue, underwritten by JPMorgan and Morgan Stanley, to replace the capital that had just evaporated. Net income for 2007 fell to €947 million against €5.22 billion the year before.

Bouton called the affair a fraud of "exceptional dimensions" and offered his resignation twice; the board refused it twice, and he gave up his salary for six months before stepping down as chief executive in 2008 and as chairman the following year. President Nicolas Sarkozy's office made it known within days that the bank should not fall into foreign hands, which was read in Paris as a warning to BNP Paribas, whose interest in a weakened competitor was not hypothetical. In July the Commission Bancaire fined Société Générale €4 million for the control failures — a sum equal to roughly one part in twelve hundred of the loss.

Who Was on Trial

Kerviel never denied the trades and never took a euro out of the bank for himself. His defence, from the first police interview to the last appeal, was that his superiors had tolerated the profits and therefore the method — that a book generating tens of millions from a desk with no capital at work was not invisible so much as unexamined. Prosecutors answered that tolerance is not authorisation and that forged emails are not an accounting oversight.

The Paris court convicted him on 5 October 2010 of breach of trust, forgery and unauthorised computer use, sentenced him to five years with two suspended, and ordered him to repay the bank €4.9 billion — a figure roughly 150,000 times his annual salary, and one no court expected to collect. He served his sentence from May 2014, having walked from Rome to Paris that spring after an audience with Pope Francis, and was released in September under an electronic tag.

The civil half of the judgment did not survive. In 2014 the Cour de cassation upheld the criminal conviction but quashed the damages, holding that the lower court had failed to weigh the bank's own failings. A Versailles appeal court reassessed the sum in September 2016 and set it at €1 million, finding that Société Générale's control lapses had contributed to its own loss. Separately, French tax law had allowed the bank to deduct the loss, a benefit estimated at about €1.7 billion, and the arithmetic of a state effectively financing a third of the damage kept the case in the National Assembly long after the courts were finished with it.

What the Loss Bought

Société Générale survived. The rights issue was taken up, the bank avoided the fate of the institutions that failed later that year in the 2008 crisis, and it spent the next three years rebuilding trade-capture and reconciliation systems around the principle the Green report had identified: that a hedge is only a hedge if someone outside the building confirms it exists.

The pattern was old by 2008. Nick Leeson's destruction of Barings had turned on the same fusion of trading and settlement authority; Yasuo Hamanaka's copper book at Sumitomo had run for a decade behind falsified confirmations; Long-Term Capital Management had shown in 1998 what happens when nobody aggregates a gross position across counterparties. Bill Hwang's family office would demonstrate in 2021 that the aggregation problem had migrated rather than been solved.

Société Générale did not close the Delta One desk; it rebuilt the controls around it. Three and a half years after Kerviel was walked out of La Défense, UBS discovered that a trader on its own Delta One desk in London had lost $2.3 billion, concealed behind fictitious forward trades with deferred settlement dates.