by Liam Vaughan
Geithner asked the participants to talk through what they’d ascertained one by one. The day’s events bore some striking similarities to 1987. Amid skittish markets, a sharp fall in S&P index futures had fed through to the stock market, causing a ramp up in trading volume as participants bolted for the exit. This frenzied activity pushed the plumbing underpinning the markets to its breaking point, leading to delays in data feeds and some wildly anomalous trades. The big difference was that, while the ’87 crash had played out over a single day (‘Black Monday’), the Flash Crash was over in half an hour. And thankfully, this time the markets had bounced back. In terms of who or what had caused it, nobody really had a clue. Rumours ranged from a fat-fingered trader to hackers to terrorists. But the world had moved on from the days when the market’s machinations could be observed by the human eye. To get anything approaching a definitive answer, the CFTC, which oversees futures, and the SEC, which is responsible for stocks, would need to obtain and analyse the trading data, microsecond by microsecond, and build a picture of who was doing what and when. It was a gargantuan task of a scale never previously attempted in the electronic era.
One might expect regulators to have real-time oversight of what goes on in their markets, but in fact data is collected by the exchanges and passed along only upon request. In the case of futures, where most products trade on a single exchange, that was relatively straightforward. The CME had already sent a file containing details of the day’s e-mini transactions to the CFTC. Stocks were a different story. Since 2005 and the introduction of a set of rules designed to increase competition, the American stock market had become chaotically fragmented. Dozens of exchanges, electronic networks and obscure trading venues had sprung up from which investors could buy shares, and the SEC would have to wait to hear back from all of them before it could carry out a complete analysis. Geithner proposed that he, Gensler and Schapiro meet with the heads of the major exchanges the following week and drew the conversation to a close.
Notwithstanding this information black hole, the government was under pressure to reassure the public that it had matters under control. Throughout the next day and into the weekend, the Flash Crash dominated the press, and the consensus was that the rise in algorithmic trading was to blame. ‘High-Speed Trading Glitch Costs Investors Billions,’ wrote the New York Times. The timing could hardly have been worse. That month, the biggest finance bill since the Great Depression, the Dodd-Frank Wall Street Reform and Consumer Protection bill, was wheedling its way across the floor of the Senate. It was a mammoth piece of legislation that already had staff at the agencies working around the clock. Its focus, however, was on regulating derivatives and making banks more resilient in light of the 2008 financial crisis. There was almost no mention of algorithms or HFT, leading some lawmakers to question whether the authorities were so busy addressing the dangers revealed by the last crisis that they’d failed to spot the iceberg looming up ahead. On 7 May the day after the crash, two Democratic senators, Ted Kaufman and Mark Warner, published a letter demanding that regulators report back to Congress on what happened within sixty days of the ratification of Dodd-Frank. ‘A temporary $1 trillion drop in market value is an unacceptable consequence of a software glitch,’ they wrote.
In response to the clamour, the government did what it often does and formed a committee – the Joint CFTC/SEC Advisory Committee on Emerging Regulation – comprising industry leaders, former regulators and Nobel Prize-winning professors. Its mandate was to consider what, if any, changes should be made to the structure and oversight of the markets in this brave new automated world. It was an illustrious group, but none had any direct experience with high-frequency trading. The youngest member was fifty-five. Meanwhile, staff at the CFTC and the SEC set to work piecing together the events of 6 May. Responsibility for the CFTC portion was given to Cyrus Amir-Mokri, an urbane former partner at the elite law firm Skadden, Arps. The goal was to get the Flash Crash investigation wrapped up as quickly as possible while demonstrating that they’d taken the matter seriously.
Amir-Mokri formed a team of around twenty attorneys, investigators and economists from across the CFTC’s nine-storey, orange-bricked headquarters in central Washington. Some, from the divisions of Enforcement and Market Oversight, were instructed to interview the most active traders on the day of the crash. Others were tasked with examining the broader macroeconomic conditions. The detailed, trade-by-trade analysis was to be handled by a group of academics led by a young, Ukrainian-born economist named Andrei Kirilenko. Each evening at 7 p.m., the team convened in a conference room on the ninth floor to discuss their progress. For days, they hardly slept. ‘It was pretty extreme,’ recalls one staffer. ‘We’d keep hearing: “Congress is calling, the White House is calling. They need answers!” ’
Adding to the pressure was a constant flow of commentary and speculation in the press. One article in the Wall Street Journal suggested, with a hint of irony, that trading by a fund advised by Nassim Nicholas Taleb, author of Black Swan, the best-selling treatise on the probability of extreme economic events, may have played a part in the meltdown. Another, on CNBC’s website, cited chatter about a mistyped trade in Procter & Gamble shares being the trigger. Then there were the victims’ stories. After markets closed on 6 May, a broker-dealer industry body called FINRA had struck a deal with the exchanges to cancel any trades that occurred more than 60 per cent away from their price before the crash started. Around twenty thousand mostly equities trades were scrapped, but not all transactions met the threshold, resulting in some big losers.
Mike McCarthy, an unemployed father of three from South Carolina, had inherited a modest stock portfolio when his mother died in 2009. As conditions deteriorated on the afternoon of the crash, he called up his broker at Morgan Stanley in a panic and told her to start liquidating his holdings, including 738 shares in Procter & Gamble. Unfortunately for him, the broker executed the order just as the market tanked, and McCarthy ended up receiving $39 per share instead of the roughly $60 he would have got if the trade had gone through either twenty minutes earlier or later. That twist of fate cost him around $17,000. ‘That’s like six-to-eight months of my mortgage payments right there,’ he told a reporter at The Street. Across the country, in Dallas, a small hedge fund called NorCap was burned when, stuck in a losing trade, it was forced to buy some options to cover its position. The order was placed just as the options spiked from around ninety cents to $30, and since FINRA’s erroneous trade agreement didn’t cover derivatives and the counterparties on the trade, including HFT giant Citadel, refused to rip up the deal, the fund lost more than $3 million.
In a turn of events that would surely have pleased Sarao, Igor Oystacher, his perceived nemesis, lost $3.5 million in a minute after loading up on e-minis after the S&P 500 had fallen 4 per cent. Oystacher correctly predicted the market would rebound, but the Russian lost his nerve as the market continued to tumble, and he exited his position for the biggest loss of his career. A few days later he quit Gelber, the firm he’d joined as a student, to start his own outfit. Another casualty was Danny Riley, an experienced pit broker whose operation was almost wiped out when one of his traders accepted a huge order from a hedge fund moments before the market bounced back. ‘We’d watched the S&P completely fall apart, there was a squawk-box guy screaming … (and after it started to rebound) I remember leaning down to look at the other end of the desk,’ Riley recalls. ‘Everybody seemed to be cool and I said “hey is everybody ok?” And everybody nodded their head except one guy at the end who shook his head that he had a problem.’ Riley pleaded with the counterparty to rip up the trade but it refused, and the firm ended up losing around $8 million. He apportions some of the blame to the regulators for failing to curtail the rise of algorithmic trading. ‘Who let this stuff in the backdoor and who hasn’t controlled it?’ he said in an interview on the Stocktwits podcast. ‘I’m a Chicago guy. Things have gotten out of control with this.’
A week after the crash
, Gensler, Schapiro and the heads of the major exchanges convened in a large, ornate meeting room near Geithner’s office at the Treasury. The atmosphere was tense. On one side, representing the futures industry, was the CME Group’s chief executive, Terry Duffy, a bombastic former pit trader of Irish stock who guarded the reputation of ‘the Merc’ like it was his own family. On the other, speaking for the securities markets, were executives from the likes of the New York Stock Exchange and Nasdaq. Tensions had been brewing since the afternoon of the crash. ‘There was just this big jurisdictional fight to say that the SEC markets were fucked-up and the CFTC markets behaved properly or vice versa,’ recalls one CFTC official. For two hours, executives took turns explaining how their market had performed exactly as it should when others had failed. It echoed a similar spat between futures and equities in 1987. By the end of the meeting, there was no love lost and little consensus on where to lay the blame.
Back at the CFTC, the investigators’ first task was to ascertain who was most active in the e-mini in the period before and immediately after 1.41 p.m. CET, the ground zero of the crash. Trawling through the trade records, they quickly discovered that one entity had sold considerably more e-minis that afternoon than anyone else: Waddell & Reed, a moderately large mutual fund founded in Kansas City in 1937 by a pair of World War I veterans to help families plan for the future ‘one step at a time … regardless of background or level of wealth’. Waddell & Reed’s flagship investment vehicle was the $27 billion Ivy Asset Strategy Fund, which was comanaged by the firm’s president, Michael Avery. For most of 2010, Avery had been bullish, and on the day of the crash, 87 per cent of the fund’s assets were in shares. But that morning, as the Dow tumbled and bad news from Europe dominated the headlines, his outlook changed and he ordered his staff to hedge their exposure to the stock market by selling seventy-five thousand e-minis worth $4.1 billion. That way, if the market kept falling, the losses the fund took on the shares would be at least partially offset by gains from the e-mini’s decline. Normally, the firm’s head trader, a man named Jeff Albright, would have overseen such a major transaction, but that day Albright and several of his colleagues were at the bar of the Hilton President hotel at an event hosted by the Kansas City Securities Association. In their absence, the traders in the office utilised an algorithmic platform provided by the British bank Barclays to carry out the trade. In and of itself that wasn’t unusual: fund managers invariably use algos to break up large orders and execute them stealthily. It was the specific program they selected that was the problem.
Barclays’s platform allowed customers to determine how they executed an order based on variables including market volume, price and speed. Waddell & Reed’s traders opted for a variant that would sell e-minis at a rate of 9 per cent of the total trading volume over the previous minute. The idea was that, as trading volumes rose and the market demonstrated greater capacity to absorb the order, selling would increase; and when the marketplace slowed down the algo would ease up. However, Waddell & Reed had failed to incorporate any kind of emergency fail-safe, such as a minimum price it was willing to accept per e-mini or a ceiling on how many contracts it was prepared to offload in a given time frame. Under normal conditions, the algo would have served Waddell & Reed’s goal of executing a substantial order under the radar, eking out e-minis slowly enough that other participants wouldn’t notice. But when they switched the program on at 1.32 p.m., after a torrid day in Europe, markets were severely out of whack. Within seven minutes, Waddell & Reed’s algo had already offloaded around fourteen thousand e-minis. By now, a number of HFTs and hedge funds had become so perturbed by conditions that they’d shut down their machines, and the few buyers who remained disappeared. At 1.41 p.m., with nothing to prop it up, the e-mini started falling like a runaway lift. If Waddell & Reed’s traders had been more responsive, they might have shut the program down. Instead, their volume-sensitive algo actually sped up since in spite of the exodus – trading activity soared. The CFTC would later conclude that, with no one else to transact with, the automated players that stuck around had gone into a kind of frenzy, firing e-minis back and forth among themselves in what the agency called a ‘hot potato’ effect. By the time the CME’s stop-logic function kicked in at 1.45 p.m. and 28 seconds, halting trading for five seconds, Waddell & Reed had sold $1.9 billion worth of e-minis. Over the next six minutes, as the market bounced back, it unloaded the remaining $2.2 billion. The entire trade, the biggest e-mini transaction of the year, had taken less than twenty minutes and it had occurred during the equivalent of a force 12 hurricane. Gensler described the execution in a speech as like ‘auto pilot(ing) into a ravine’.
The CFTC was in no doubt that Waddell & Reed’s mammoth sell order was a major spark for the Flash Crash, but Kirilenko, the economist, wanted to know what role HFT had played. After all, funds and banks had been placing ill-timed and ill-conceived trades forever, and yet markets had never plummeted with such ferocity. Since the preceding year, Kirilenko and his colleagues in a department called the Office of the Chief Economist, many of them PhD students lured to the agency by the prospect of getting access to up-to-the-minute, confidential trade data, had been building a kind of typography of modern electronic futures markets in an effort to understand how pervasive HFT had become, and why it was so profitable. After the crash, that work came to the fore. ‘HFT was still pretty low on the agenda,’ recalls Albert ‘Pete’ Kyle, a University of Maryland professor who was at the CFTC at the time, ‘but we had all these questions, like “What would happen if one of these algos blew up?” and “Do they promote or interfere with competitive market-making?” ’
On 6 May, the researchers found, there were a total of around fifteen thousand participants active in the e-mini, ranging from the smallest one-lot rookies at the likes of Futex to the biggest global pension funds. Of these, they classified sixteen as HFT based on their attributes of trading huge volumes without ever accumulating a large position, and always ending the day close to flat. Despite making up just 0.001 per cent of the participants in the trading universe, this group of largely Chicago-based entities was responsible for 29 per cent of the trading volume. The HFT lobby liked to talk about how they brought liquidity to the market, making it easier for all participants to buy and sell at reasonable prices whenever they wanted to; but the data suggested that the most profitable of these firms mostly removed liquidity by aggressively hitting the resting offers of other participants just as the market was about to move, something academics refer to as ‘sniping’. And on the afternoon of the crash, the majority of HFT firms had either shut down or joined the selling frenzy, pushing prices down further. ‘During the Flash Crash, the trading behaviour of HFTs appears to have exacerbated the downward move in prices,’ Kirilenko and his colleagues wrote. ‘We believe that technological innovation is essential for market advancement. As markets advance, however, safeguards must be appropriately adjusted to preserve the integrity of financial markets.’
For its part, by late summer the SEC still hadn’t managed to corral the nation’s equity markets to hand over sufficient trading data to allow it to carry out the kind of analysis for stocks that the CFTC had managed for futures. Absent the numbers, the agency resorted to interviewing traders about their experiences. The conversations suggested there were actually two crashes that afternoon: one in the e-mini between 1.41 p.m. and 1.45 p.m., and a second in individual stocks from 1.45 p.m. until around 2 p.m. In a sign of the highly interconnected nature of modern markets, when the e-mini collapsed, it triggered stock traders’ systems to shut down automatically. This, in turn, precipitated a bout of selling and, with so few buyers left, shares cascaded. In a few minutes, more than one thousand stocks and exchange-traded funds fell by at least 10 per cent. Fuelling the exodus was a technical issue at the biggest stock market of all, the New York Stock Exchange, which happened to be upgrading its IT system that day and ended up quoting prices up to twenty seconds old – a virtual lifetime in a marketplace whe
re speed was measured in millionths of a second.
The mystery of why some shares changed hands for less than a cent and others for $100,000 came down to an arcane rule that obliged market makers in a given stock to provide quotes at all times, regardless of conditions. The stipulation was supposed to help prevent events like the Flash Crash by guaranteeing there was always somebody willing to trade, but, as ever on Wall Street, some dealers had found a way around it by leaving extremely high or low quotes in the order book. Before 6 May 2010, it had seemed inconceivable that these so-called ‘stub quotes’ would ever be hit.
On 30 September, nearly five months after the crash, the CFTC and the SEC published their joint paper. Determining what caused a market crash is a bit like saying what caused World War I or the rise of Hitler. In a highly complex system, there are always a multitude of factors to which different people will attribute different weights depending on how they view the world. The picture the regulators painted was of a perfect storm in which a huge, clumsy, one-way sell order from an old-school fund arrived at exactly the wrong time, sending an already highly volatile market into meltdown. HFT firms hadn’t caused the crash, the authors concluded, but, by either turbocharging the selling or running for the exit, they certainly hadn’t helped matters either.
The conclusions inevitably drew criticism, not least from Waddell & Reed, which, backed up by the CME Group, put out a statement saying there was ‘no evidence to suggest that our trades disrupted the market on May 6’ and ‘trades of the size we initiated normally are absorbed easily by the market’. Dave Cummings, an HFT pioneer and the founder of a firm called Tradebot Systems, disagreed. ‘Who puts in a $4.1bn order without a limit price?’ he wrote in a scathing email picked up in the press. ‘It angers me when people blame technology for what are clearly lapses in human judgment.’