Preventing a repeat of the flash crash

A report by US regulators offers a post-mortem on the May 6 flash crash and gives clues as to how a future stock market dive could be prevented. But not all market participants agree with the findings – or the possible fixes. Mark Pengelly reports

The flash crash might have been precipitated in minutes, but regulators and market participants have spent months turning over the details of exactly what happened. The basic facts are clear: at around 2:45pm on May 6, the Dow Jones Industrial Average (DJIA) went into freefall, quickly collapsing by 998.5 points before rebounding almost as rapidly. The stock index eventually closed at 10,520.32 points, down just 322.86 points on the day.

Speculation over the causes of the carnage is rife. Some have suggested a rogue execution algorithm caused the crash, while others have put the blame on a series of dud trades in certain stocks. Another theory centres on an apparent volley of quotes by high-frequency trading firms, which caused delays in certain markets. In fact, the answer could involve all of the above - at least, so says a report into the crash published by the US Commodity Futures Trading Commission and the Securities and Exchange Commission (SEC) last month.

The report, which was released on October 1, says the flash crash was caused by an array of different factors - something that will frustrate market participants looking for easy answers. "A lot happened on that day, but I don't like to describe it as a confluence of events - it was more a chain of events. That is quite different: one thing led to the next. Those steps help highlight and illustrate some weaknesses in the current market structure for both futures and equities," says Gregg Berman, Washington, DC-based senior policy adviser at the SEC, who led the agency's investigation.

May 6 was shaping up to be a turbulent day even before the flash crash occurred. The eurozone sovereign debt crisis was at its peak, with plans to slash public spending in Greece leading to protests and riots. Other European states - in particular, Ireland, Italy, Portugal and Spain - also looked to be under pressure, with fears of contagion throughout the continent starting to build.

By 2:30pm, the DJIA was down 2.5% from its opening level and the Chicago Board Options Exchange's Vix index, which reflects 30-day implied volatility as measured by S&P 500 option prices, was up 22.5%. Liquidity also fell in two of the most active stock index instruments - the E-mini S&P 500 futures contract and the S&P 500 Standard & Poor's Depository Receipts (SPDR) exchange-traded fund (ETF). Buy-side liquidity in the E-mini declined from $6 billion in early morning trading to $2.65 billion, while liquidity in the S&P 500 SPDR ETF fell from about $275 million to $220 million over the same period, the report found. Individual stocks were suffering from similar declines in liquidity, with many hitting liquidity replenishment points (LRPs) on the New York Stock Exchange (NYSE). LRPs are designed to slow trading in stocks by preventing electronic execution during periods of high volatility.

Into this environment strayed an execution algorithm, used by a "large mutual fund complex" to sell 75,000 E-mini futures contracts as a hedge for an existing position, the report says. The fund is not identified in the report for legal reasons, but is widely believed to be Kansas City-based Waddell & Reed. The firm would not comment directly on the matter. The sell order, which comprised 75,000 June 2010 E-mini contracts valued at around $4.1 billion, resulted in the biggest net change in daily position of any trader in the contract since the beginning of the year, according to the report. The algorithm was designed to target an execution rate set to 9% of trading volume calculated over the previous minute, but crucially did not take price or time into account.

"The way the algorithm sold dynamically changed the behaviour of other market participants. As that happened, it caused a liquidity issue and in turn that caused the pricing of the E-mini to plummet particularly fast," explains Berman.

To put it in perspective, only two single-day sell programs of similar size had been executed in the 12 months before May 6. One of those was conducted by the same firm, but on that occasion it took more than five hours to complete the sale of the contracts. This time, it occurred in just 20 minutes, gyrating the futures market and causing a knock-on effect in equities.

Although the selling was initially opposed by buying interest from high-frequency traders and other market participants, this interest began to stall as their long positions grew. At this point, they also became heavy sellers to reduce their exposures. In turn, the algorithm responded by increasing the rate at which it sold, causing liquidity to dry up.

In the E-mini, the absence of some market participants created a ‘hot potato' effect, as high-frequency traders rapidly bought and sold contracts with each other. Meanwhile, in equity markets, investors began to place large numbers of sell orders for individual stocks and ETFs. Buying interest disappeared as market participants withdrew, worried the dislocation could be a data error or the result of an event they did not know about. Even market-makers became frightened. Instead of the usual practice of offsetting large numbers of buy and sell orders internally, dealers began routing bulky sell orders into the public market, exacerbating the distress.

"When you had these very large rapid moves, market-makers elected to pull back and reduce their rate of internalisation and a few stopped processing altogether. They are a source of liquidity, and when they stopped doing that, it also contributed to the fall in liquidity," says Berman.

The evaporation of liquidity caused some stocks to be traded at stub quotes - essentially, bids and offers provided by market-makers that are unrealistically high or low, meant to fulfil their obligations to provide continuous two-way quotes, even though they have withdrawn from active trading. That led to a number of irrational trades, with some stocks, such as those of consulting company Accenture, trading at just a penny.

When it comes to the role played by high-frequency traders, the report is ambiguous. Fingers were pointed at these participants in the aftermath of the crash, with some claiming a bombardment of quotes by high-speed traders had led to the sudden market slide - a theory that seems to be backed up by analysis from Chicago-based software vendor Nanex1,2. Some observers have gone even further, suggesting high-frequency traders are guilty of so-called quote stuffing, which involves these firms placing large numbers of orders before quickly cancelling them to overload exchange systems and gain an edge over other market participants.

"People came up with the theory that not only did the data delays cause the crash, but the data delays were caused by too many quotes coming in. And those quotes were being purposely sent into the market to cause data delays, such that the person doing that would be able to somehow profit by doing that. That's a lot of ifs," says Berman.

On May 6, the consolidated feed provided by the NYSE lagged while its proprietary feed kept in line with other markets, the report says. However, Berman doesn't believe quote-stuffing was at play. For one thing, market participants interviewed by regulators say they don't use the consolidated feed for making trading decisions, he says. "When we talked to people about their use of the consolidated tape, most of the liquidity providers said actually they don't use the consolidated tape to make trading decisions - they use the proprietary feeds."

Berman thinks it is unfortunate some observers have chosen to jump to their own conclusions about the role played by high-frequency traders on May 6. The report, drawn up by a team of between 20 and 30 staff at the SEC alone, includes data others do not have access to and has been submitted for peer review, he says. Significantly, it also draws upon three months of interviews with individual market participants. "If you had to do just a statistical analysis, you would only be able to go so far in terms of your analysis. You would be able to understand what happened, but you wouldn't be able to understand exactly why it happened. We are very fortunate in that we were able to reach out and speak to many participants about that day."

Despite this, the report notes the SEC will work with market centres "in exploring their members' trading practices to identify any unintentional or potentially abusive or manipulative conduct" that may cause system delays.

Proponents of high-frequency trading say these firms supply markets with much-needed liquidity, reducing volatility and resulting in lower bid-offer spreads for other investors.

"High-frequency trading has had a very positive effect, both in terms of liquidity and narrowing of spreads, which is positive for all investors - both institutional and retail. So anyone trying to blame high-frequency trading for the crash is off-the-mark," asserts John Bohan, New York-based head of cash equities at BNP Paribas.

Nevertheless, the joint report shows this is not necessarily the case. In fact, high-frequency traders were forced to step back after their initial buying activity on May 6, says Berman. "What we found is that over the course of the day, high-frequency traders took as much liquidity as they provided. They were taking liquidity and providing liquidity on both the sell and buy side. When there was a big wave of selling, they seemed to have ridden the wave along with many other market participants. The data did not suggest they stabilised prices during the sell-off," he says.

How might another flash crash be prevented? The role of the E-mini trades that preceded the crash has been a focus for some market onlookers. The algorithm that executed the E-mini sales was"clearly stupid and behaved poorly in the way it executed", says one New York-based electronic trading head at a major dealer.

Andrew Lo, professor of finance at the MIT Sloan School of Management in Massachusetts, agrees. "When the algorithm said trade 9% of the trading volume over the past few minutes, it turned out that was a really big number. It was like ramming a huge amount of food down somebody's throat," he says.

A research note by New York-based agency broker Instinet on October 14 asserts the first lesson from the flash crash is not to use execution algorithms that do not contain price controls. This applies to the algorithm employed by the mutual fund, as well as the algorithms used by other high-frequency traders. "If the algorithms contributing to the problem on May 6 had been more sensitive to market conditions and aware of the type of volume being traded, they would likely not have been so aggressive," the paper says.

According to sources with knowledge of the matter, the algorithm was a standard participation type used by brokers in the futures market. Given their widespread use, it would be impossible for regulators to comb through countless algorithms in search of problems, says Lo. Other market observers believe brokers should bear more of the responsibility to ensure client orders are appropriately executed.

"You have to put the onus on the broker to monitor client behaviour. The issue is whether, under any reasonable monitoring guidelines, the order would have gone through. That is open to debate," says Adam Sussman, New York-based director of research at Tabb Group. An idea touted by some is the outright elimination of market orders - or orders intended to be executed at the prevailing market price, whatever that may be.

Nonetheless, the significance of the algorithm in question has been hotly disputed (see below). Instead, regulators have so far used the crash to address broader issues of market structure - an approach that is positive, believes MIT's Lo. "Markets have evolved more rapidly than our regulatory infrastructure. As a result, these situations are more common now because there is a mismatch," he says.

An obvious place to look for improvements is the SEC's circuit-breaker regime, thinks Lo. The SEC already had market-wide circuit breakers in place on May 6, which initiate a trading pause when stocks experience significant volatility. However, none of the circuit breakers was triggered by the crash. "The circuit breakers already in place did not work and they need to be modernised. If your circuit breaker does not trigger at home and you have an electrical fire and your house burns down, obviously there's something wrong with those circuit breakers," says Lo.

Berman at the SEC agrees. One of the important conclusions in the report is that market-makers will occasionally step back and pause during the course of trading. Better managing those pauses is a responsibility of regulators, he suggests. "Under certain circumstances, such as those on May 6, market-makers will pause their trading, will pull back and will reduce the liquidity. Given that's the way the markets seem to operate, we need to have protections against trading in that kind of environment," he says.

At the moment, market-wide trading halts are called once the DJIA suffers a 10%, 20% or 30% fall, as calculated from the beginning of every quarter. The SEC wants to extend these rules, and imposed a pilot circuit-breaker regime for individual stocks in June. Under these rules, stocks that experience a 10% change in price over a five-minute period will stop trading for five minutes. Having been initially applied to equities in the S&P 500 index, they were extended to cover the Russell 1000 index and some ETFs on September 10.

 

Limit up/limit down

Another improvement being considered is the imposition of a so-called limit up/limit down regime, which is already used in the futures market. This would effectively impose price bands, restricting the maximum possible movement of a stock during the trading day. "I think there is a general sense in the industry that some sort of a limit up/limit down banding of prices would dramatically reduce the chance of this type of event from happening again," says Berman.

The supervisor is also reviewing the criteria under which erroneous trades are broken. During the crash, some market-makers were scared away by uncertainty over which trades would be broken and which would continue to stand, the report says. On May 6, exchanges only broke trades that were more than 60% away from the applicable reference price, and did so using a process that was not transparent, the report found.

Jose Marques, global head of electronic equity trading at Deutsche Bank in New York, says this kind of situation makes it difficult to make two-way markets, since dealers are unaware of what their true risk position is. In light of the crash, greater certainty is essential, he adds. "The market needs stability. We can't have markets, whether they are electronic or human-driven, in which stocks can trade at crazy prices and then be busted 10 minutes later. That's inherently disorderly."

The SEC approved new standards for broken trades on September 10, making it clearer which transactions would be broken by exchanges. For stocks covered by the individual circuit-breaker programme, trades will be broken if they are 10%, 5% or 3% away from the circuit-breaker trigger price, depending on whether they are trading at under $25, between $25 and $50 or more than $50. For stocks not covered by circuit breakers, deals will be broken depending on how many equities are involved. For events involving between five and 20 stocks, trades will be broken if they are at least 10% away from the last price before trading was disrupted. For events involving more than 20 stocks, trades will be broken if they are at least 30% away from this price.

Alongside these rules, tighter circuit breakers should also provide a form of pre-screening for transactions that are likely to get broken, says Deutsche Bank's Marques.

Meanwhile, banning stub quotes - another measure being looked at by the SEC - would help prevent trades from being executed at absurd prices in the event of another market meltdown. "Eliminating stub quotes is the ultimate no-brainer," comments Paul Zubulake, senior analyst at consultancy Aite Group in New York.

Although it was arguably not a direct cause of the crash, many market participants think the highly fragmented structure of the US equity market is also partly to blame for what happened. "There's tremendous fragmentation in the market. Although that is good for competition, it means there are no harmonised or consistent rules across the different exchanges. Frankly, I think that is the biggest cause of the volatility we saw on May 6," says Bohan at BNP Paribas.

Inconsistencies in broken trades are one example of this, he says, but another is the use of mechanisms designed to slow trading in volatile markets. Bohan says the triggering of LRPs on the NYSE and the use of Nasdaq's volatility guard - a system of circuit breakers unique to the exchange - had a detrimental effect during the crash. "The existence of different rules across exchanges, such as the NYSE's LRPs and Nasdaq's volatility guard, caused orders to be routed to other venues for instant execution at potentially worse prices," he explains. The joint report says LRPs were not to blame for the flash crash, but acknowledges they did influence the decision of many participants to step back from the market. A harmonisation of exchange rules would leave the market better able to deal with future high-speed dislocations, many argue.

 

A flawed execution

The significance of the execution algorithm apparently used by Waddell & Reed shortly before the flash crash has been questioned by the Kansas City-based mutual fund and other market participants. According to the joint report by the US Commodity Futures Trading Commission and the Securities and Exchange Commission (SEC), the algorithm’s use – and in particular, its lack of regard for price or time – played a pivotal part in the events of May 6.

“Following the recent release of the regulatory report on the flash crash, many market observers have noted the events of May 6 involve multiple issues that transcend the actions of any single market participant. We agree with those observations,” said Waddell & Reed in a statement.

Meanwhile, Chicago-based software vendor Nanex – a company that maintains a proprietary database of archived quote and trade information for all US equity, options and futures exchanges – has used the original trade executions from the firm to refute the findings of regulators. “Our analysis doesn’t match the description of an algorithm engaged in indiscriminate selling,” says Eric Hunsader, software developer and president at the firm.

The firm matched the 6,438 original trades by price, size and time, to the 147,577 trades in the Chicago Mercantile Exchange’s time and sales data between 2:32pm and 2:52pm on May 6. The figures show the bulk of the algorithm’s selling activity did not occur during the market dive, he says. “The Waddell & Reed algorithm really started executing heavily after the market bottomed. Then it took over and sold a significant amount on a rally,” he says. The algorithm’s activity proves it was sensitive to price, he says, which prevented its selling activity from continuing as markets underwent a precipitous drop.

Gregg Berman, Washington DC-based senior policy adviser at the SEC, who led its inquiry into the flash crash, admits much of the algorithm’s selling activity occurred after the market had bottomed. However, he maintains it was not sensitive to price or time.

1 www.risk.net/1730959
2 www.risk.net/1736728

A version of this article appeared in Risk magazine