ParisTech Review

On May 6, at around 2:30, the Dow Jones Industrial Average began to drop – like a very heavy stone. In less than 15 minutes, it lost more than 8 percent of its value, plummeting to 9,870, from 10,700. Shares in Accenture, the global consultancy, and several other stocks fell briefly to 1 cent. At its trough, nearly $1 trillion in market value had disappeared. At 2:45, the market began to right itself, and closed just a few hundred billion dollars down.

The mysterious “flash crash” set off alarms throughout the global regulatory community. One prime suspect: “black box” trading systems, the computer programs that now buy and sell a major share of the world’s liquid investments with no human intervention. It’s unclear yet what role, if any, the robo-traders played in the crash, but the shock of the event has served to make the public more aware of how much Wall Street has changed in just the past few years. In particular, it has market experts debating anew over the level of risk such trading systems are adding to the markets – and how uncontrollable those risks may be.

Rise of the Algorithm
In the old days, investors telephoned brokers who telephoned someone on the floor of a stock exchange or a commodities trading pit to execute an order. Over the past two decades, however, that began to change, as human beings were thrown off most trading floors and replaced by computerized matching engines. Trading continued, of course, but the loud men in the louder jackets were replaced by quieter people at desks who were supported by analytical software.

Now, human traders are in the process of being logged out altogether as computers are replacing not only the physical markets but the people manning the desks in the electronic markets. In the United States, computers now buy and sell 66 percent of all stocks and 25 to 30 percent of futures and derivatives contracts, according to the TABB Group, a financial services consultancy, and markets worldwide are making similar transitions.

Computerized trading has been around since the 1980s, and is sometimes blamed for the stock market crash of 1987. In recent years, however, traders have used it more and more, as they learned to take advantage of the computer’s special abilities to make decisions at speeds thousands of times faster than any human being ever could.

The black boxes buy and sell on preprogrammed instruction, all without the assistance of any human intervention. Physically located near the matching engines of the electronic exchanges for ultra-high-speed access, these algorithmic programs are designed to hunt for tiny variations in prices of orders and correlations between markets.

It’s a continuation of a trend that’s been around for as long as the stock markets have been around, according to Thierry Foucault, a professor of finance at HEC, the ParisTech business school. It’s no different, he says, than when brokers bought seats on the stock exchange floor. “The reason why people were willing to buy a space on the floor was simply that they could access information more quickly.”

To meet the demand, the exchanges have had to pick up the pace as well. At Eurex, for instance, in Frankfurt, orders have sped up tremendously, according to Deutsche Börse Systems, Eurex’s IT provider and a wholly-owned subsidiary of Deutsche Börse, one of Eurex’s primary shareholders. Four years ago, it took several hundred milliseconds to make a transaction in the exchange, executives say. Now, Deutsche Börse engineers have tweaked it so the order takes between 1.5 and 2 milliseconds.

Algorithmic trading comes in two basic flavors. The first involves high-frequency traders who act essentially as market makers, buying and selling on each up or down tick in the market. At Eurex, for instance, traditional, human-operated market-making desks have been replaced largely by computers. The second flavor consists of traders who use a variety of strategies – some high frequency, some not – that mostly try to profit from slight pricing differences that pop up in the relationship of certain assets, such as the same stock traded in different markets or different stocks and securities that normally trade within a certain range of each other.

Harrell Smith, head of product strategy for Portware, a New York company that produces a software platform used by algorithmic traders, says that in the futures markets, at least, he expects the share of black-box trades to continue to climb. “There’s really no reason why it shouldn’t,” he says.

Until a few years ago, major banks designed most of these systems, for their own proprietary trading. Today, the software is available at a nearly industrial scale, observers say. To form a trading company these days, you need just a handful of mathematicians and software programmers plus some off-the-shelf software like Portware. Even capital is not much of an obstacle, given that ownership may last for no more than a few seconds.

The growth of algorithmic trading may be changing the market in another way, too, by reducing the need for brokers. “In the past, the broker was useful in terms of information, and in terms of implementation. But increasingly clients can get data feeds in real time from Reuters, Bloomberg or the [trading] platform and they can start their own trading desk,” says HEC’s Foucault.

Most analysts say these new programs seem to be increasing market liquidity, making it easier for sellers to find buyers regardless of whether a stock is headed up or down. “Ultimately, for the most part, this is providing considerable liquidity to the markets and making the markets more efficient,” says Kevin McPartland, a senior analyst at TABB in New York.

Risk and Reward
Manufacturers and operators of such systems say that the black boxes are actually better than human traders because they make investment decisions unclouded by emotion.

Critics fear, however, that this same focus may also have some unintended consequences if the programs all buy and sell on the same signals.

Old-fashioned market makers were obliged to stay in the game regardless of market conditions, but today’s market makers can get out at any time, notes Steve Ohana, a Paris-based risk management expert and finance professor at ESCP Europe.

This can have some disconcerting results in the event of a sudden shock. High-frequency traders may suddenly exit declining markets, because they are programmed not to buy declining stocks. So, instead of stabilizing as a human-driven market would as investors move in to hunt for bargains, a black-box market could well go into free fall. “Automatically, the liquidity dries up and amplifies the original move that triggered the decline,” Ohana says. “The magnitude of the chain reaction is something that we are completely unable to properly anticipate.”

“It’s an emerging risk,” Ohana adds, and an exceedingly complex one. “There are very few models or methodology concepts to deal with that . . . We know that a lot of algorithms interact with each other but we don’t know in exactly what way.”

Such correlations are becoming more dangerous now that the markets correlate so closely with each other, Ohana says: “I think we have gone too far, much too far in the computerization of finance. We are completely lost. We cannot control any more the monster that we have created.”

Some politicians also fear the new technology. Following the May “flash crash,” there were calls in the U.S. Congress for more regulation, particularly as rumors flew that the crash was caused by improper coding. “A temporary $1 trillion drop in market value is an unacceptable consequence of a software glitch,” said Sen. Ted Kaufman, Democrat of Delaware, and Sen. Mark Warner, Democrat of Virginia, in a joint letter to Senate Banking Committee Chairman Chris Dodd, Democrat of Connecticut. Politicians and regulators are proposing mandatory slowdowns in the event of a sudden drop in trading activity – what Paul Jorion, a French economics writer, calls “an air pocket.” The exchanges themselves are also experimenting with controlled slowdowns.

But one high-speed trading fund owner, Rishi Narang of Telesis Capital, argues that the real problem is not the black boxes per se but rather two other innovations in the system – the increase in trading platforms outside the exchanges and the growth of index trading.

Not that long ago, stocks were all traded on a few major exchanges. Now, there are many more alternative venues, including some thinly traded electronic communications networks (ECNs), says Narang, founding principal of Telesis, an investment management firm based in Marina del Rey, California, and author of Inside the Black Box (Wiley, 2009).

At the same time that the markets have fragmented, the growth of index trading, which features multiple instruments all focused on the same assets, is ensuring that assets are more and more closely correlated. The end result is that each stock becomes more susceptible to a sudden motion in one of the ECNs, and that motion then jogs the indexes in which it is tracked and the other stocks in that index, according to Narang.

“The reality is that when someone comes in and tweaks one of [these stocks or other assets], the rest will have to move . . . they are all tracking the same asset,” Narang says.

Limits to Growth
Eurex professionals argue that there is an inherent limit to high-frequency trading, just as there is a limit to market making in any market.

As for other kinds of computer-assisted trading, the only limit may be programmers’ ingenuity.

A small but growing number of investors and hedge funds, for example, are adding artificial intelligence to their black boxes. The systems they use are designed not only to execute orders based on algorithm-driven instructions but also to learn from the success or failure of those orders and to adapt accordingly.

Some other systems are even being taught how to profit from human foibles: a number are learning to monitor social media for excitement about a company. TABB’s McPartland has also heard about a group that is using ex-CIA experts to train a computer to analyze the voices of CEOs during important public pronouncements in order to determine whether they’re lying – and then trade on that insight.