NEW YORK —The irony of lifeless computer trading is that it's meant to provide a fair, emotion-free trading platform—but that may be true only when the market itself is trading in a fair, emotion-free way.
When the stock market is in panic mode, ultra-fast computer systems can't help keep prices fair, and they have fewer incentives to stand in the way of a falling market.
"It's all math-driven," said Eric Bernstein, chief operating officer at Sophis, a provider of trading and risk management software. "But in a situation where there are massive gaps in the market, it creates a bit of havoc with the program."
That was borne out Thursday, when the Dow Jones Industrial Average plunged more than 600 points in less than 15 minutes, baffling and unnerving investors. Amid the chaos, algorithmic trading shops, or algos, that rely on rapid, computer-based automated trading in theory could have provided much-needed liquidity. Instead, they stayed largely on the sidelines.
High-frequency trading platforms may provide the best bid or best offer at a specific time, but they don't necessarily provide liquidity "depth," said Peter Kenny, managing director at Knight Equity Markets. In a fast-moving market, they may not provide the best second bid or offer, or any at all.
And even those traders who waded into the selloff to provide liquidity may have found themselves penalized as a result. Trades subsequently deemed erroneous were later broken, potentially creating money-losing positions at high-frequency trading shops.
Late Thursday, the exchanges decided to cancel all trades involving swings greater than 60% of a stock's consolidated 2:40 p.m. price. That erased all orders made when a stock dipped to irrationally low prices. But it didn't erase the trades that occurred when the market rebounded. The New York Stock Exchange said that about 4,000 trades were broken Thursday after being identified as clearly erroneous as per exchange rules.
A trader who stepped in and bought a stock that was tumbling could be left unexpectedly owing shares in what is known as a short position. For example, if a stock was trading at $60 a share, but was sold for $10, the trader who bought it at $10 and then sold it later for $50 when the market rebounded would be left holding the shares short at $50, with his previous profit wiped out.
That put those who might have stepped into mitigate the freefall in a bind and left the market open to its rapid descent.
"If in fact they're responding to a true economic disaster, they will buy stock and it will continue to go lower. But if they're responding to a mistake or an overreaction and they buy, they mitigate the freefall," but end up getting penalized for it, said Dick Rosenblatt, chief executive of Rosenblatt Securities. "In this case, the stock rallies and they sell out their position for what they believe to be a profit. In fact after the trade is broken, they end up with a short position and they will again lose money. In either case, in a fully automated market, we have disincentivized liquidity providers from entering the market to limit volatility when we need them most."
Even without broken trades, algos can struggle when prices go awry. For example, a client looking to sell 1,000 shares of a certain stock at the best available price may be able to start selling at the closest bid price of, say, $60. But if that bid can't accommodate the entire 1,000-share market order, the electronic trading system will look to sell the balance of the order at the next best available bid. If that bid happens to be below $40, so be it.
To avoid selling at that sub-$40 price, some high-frequency trading shops made a choice to halt activity. "The reason they have to halt is to establish and see where there is a fair level to print these sells because there are no buyers," Sophis' Mr. Bernstein said.
In the moment, some high-frequency shops turned instead to human judgment. But since those decisions weren't coordinated across all trading platforms, the momentary drain of liquidity may have helped prevent "erroneous trades" at some shops but exacerbated the problem at others. And it left some wondering if the market is adequately structured to stop severe tumbles.
"How have we incented algorithmic traders and high-frequency traders to enter our markets in times of stress?" Mr. Rosenblatt asked. "We really haven't."
