骑士交易危机不会绝迹

    围绕骑士资本公司（Knight Capital）和其造成4.40亿美元损失的软件故障一直有一个疑问：为什么身为美国首屈一指的造市商，处理全美超过10%股票交易的骑士会把一款容易出故障的软件推向市场？

    骑士CEO托马斯•乔伊斯在一次电视采访中解释称，公司新的软件程序向市场发出了数千错误交易，是源于“一个严重的程序错误”。上周三，骑士新的软件程序与纽约证券交易所的新平台同时上线，允许像骑士这样的造市商给零售投资者提供略有折扣的股价。

    问题不在于骑士的离经叛道——过去15年，它的声誉已得到改善。问题是，电脑程序员们称，交易软件永远都做不到100%完美，特别是当一个新软件推向像股票市场这样复杂的系统时。主导当今股市的电子交易所具有的这一特性无疑将在未来引发更多的危机。

    “因为软件有很多行代码，很多变量，加之软件应用所处的不同状态，这个数字会极其庞大——数值可达10的100方，”卡耐基梅隆大学（Carnegie Mellon University）电气和计算机工程系主任埃德•施莱辛格表示。他说，仔细核对行代码不是不可能。数学定理可以进行测试。但不可能实现的一点是，评判一家公司的软件——以骑士的软件为例——将如何与数以千计的迅速流进和流出这个市场的其他算法相互作用。这只能在软件实际运用时才会发生。

    施莱辛格对完美交易软件的想法报以一笑。专家和政客们纷纷呼吁要对交易软件实施更高的标准，倡导零错误代码。对此，他回应称这是不可能的。这就像要求橄榄球外接手能在每次对抗性防守训练中都要接住橄榄球，否则就只能坐在替补席上。

    对于骑士而言，不幸的是过去15年来它赖以成功（这种成功推动了它的规模不断增长）的因素上周酿成了巨亏。与其他发现问题就立即停止交易的高频交易商不同，骑士这样的造市商运用算法将来自Scottrade、先锋公司（Vanguard）等不同客户的买卖指令进行撮合，除非有重大错误出现，就必须接单。因此，他们在交易了数以千计的股票后才意识到了问题。

    “造市商需要一直留在市场中，”柏鲁克学院（Baruch College）和纽约大学（New York University）斯特恩学院（Stern School）金融市场信息系统教授伯纳德•丹尼弗表示。因此，市场波动对造市商的影响超出快速进出市场的高频交易商。“如果我是其他人，我就可以这么说，‘嗯，让我暂停一会交易，’”丹尼弗说。“他们没有一家造市商的指令量。”

    丹尼弗说，监管机构不应追求完美的程序，而应当将重点放在程序算法失控时如何将损害降至最低：比方说，取消30分钟时间窗口内的所有交易。实际上，这种观点的背后是一些人呼吁在纽约证交所（New York Stock Exchange，简称NYSE）恢复人类造市商。人类的速度比电脑慢很多，更有可能在不正常现象升级至难以掌控之前发现苗头。但回复到纽交所人类造市商排他、垄断的交易模式，会让二十年来的进展付之东流。这根本不可能发生。丹尼弗说：“如今，我们获得了最低的买卖价差、最高的流动性和最快速的交易——对于大多数人，这是一个相当不错的市场。”

    骑士这样的造市商可能偶尔会出问题，交易可能不得不取消。但这就是当今高速电子市场必须付出的代价，对于大多数人来说，这样的代价可能还算公平。

    译者：早稻米

    One question keeps arising in the saga of Knight Capital and its $440 million software glitch: why did Knight, one of the premier U.S. market makers that handles more than 10% of total stock trading, introduce glitchy software into the market?

    CEO Thomas Joyce explained in a television interview that the company's new software program sent thousands of erroneous trades into the market because of "a large bug." This was software Knight introduced Wednesday in conjunction with the New York Stock Exchange's new platform that allows market makers like Knight (KCG) to offer slightly discounted stock prices to retail investors.

    The problem isn't that Knight is a renegade operator -- its reputation has improved over the past 15 years. The problem is that trading software can never be 100% perfect, say computer programmers, especially when it's new and introduced into a complex system like the stock market. That aspect of electronic trading, which dominates today's stock market, will certainly cause more crises in the near future.

    "Because it has so many lines of code and so many variables, the number of different states that software can be in can become [an] enormous number --10 to the 100th power," says Ed Schlesinger, head of Carnegie Mellon University's electrical and computer engineering department. Carefully reviewing lines of software code isn't impossible, he says. Math theorems can do the testing. What is impossible, though, is judging how one firm's software -- Knight's, for example -- will interact with the thousands of other algorithms rushing into and out of the market. That can only occur when the software is in actual use.

    Schlesinger chuckles at the notion of perfect trading software. Pundits and politicians have called for more stringent standards of trading programs, essentially advocating for bug-proof code. That, he responds, is impossible. It would be akin to sidelining a wide receiver until he catches every football in every practice against every opposing defensive scheme.

    Unfortunately for Knight, what made it so successful over the past 15 years -- its growing size -- worked against it this week. Unlike other high-frequency traders, who can flip the off-switch the instant they sense trouble, market makers like Knight, which use algorithms to match buyers and sellers from diverse clients such as Scottrade and Vanguard, are required to stay open for business unless there's a major error. They can trade thousands of shares before waking up to the problem.

    "Market makers are required to stay in the market," says Bernard Donefer, a professor of information systems in financial markets at Baruch College and New York University's Stern School. Volatility, therefore, affects market makers more than it does the high-frequency-traders whizzing in and out of stocks all day. "If I'm someone else, I can just say, 'Hmm, let me stop trading for a while,'" says Donefer. "They don't have nearly the amount of orders that a market maker does."

    Instead of pursuing perfect programs, Donefer says, regulators should focus on minimizing the damage when the program's algorithms go haywire; say, cancel all trades within a 30-minute window. In fact, that sentiment is behind the calls of some to reinstate human market makers on the New York Stock Exchange. Humans move much slower than computers, and are more likely to sniff out irregularities before they metastasize to unmanageable levels. But to revert to the clubby, monopolistic days of human NYSE market makers would be to throw away two decades of improvement. And it's simply not going to happen. "Right now, you got the lowest spreads, the most liquidity, the fastest trading -- it's a pretty good market for most people," says Donefer.

    Market makers like Knight may occasionally blow up, and trades may have to be cancelled. But that's the cost of today's high-speed electronic markets, and, for most people, it's probably a fair trade off.