“海島市”油輪之前以波斯灣的巖石島嶼Umm al-Maradim命名，意思是巖石之母。這個名字盡管不太有詩意，卻很恰當(dāng)。在1987年6月的下半月，它和科威特的許多其他船只都被改名，以新的名字在美國國旗之下航行，以此阻止好戰(zhàn)的伊朗對波斯灣的船只發(fā)動威脅襲擊。

然而，此舉并未取得效果，1987年10月16日周五，伊朗發(fā)動了攻擊，從法奧半島發(fā)射了蠶式導(dǎo)彈，襲擊了“海島市”的艦橋和船員居住區(qū)。16名船員受傷，船長和一名瞭望員失明。兩天后，美國海軍予以了回?fù)?，“靈敏射手行動”摧毀了伊朗的兩個用于監(jiān)視、充當(dāng)直升機(jī)和兩棲巡邏艦基地的鉆油平臺。1987年10月19日周一，美國人一早醒來，想著我們終于與伊朗開戰(zhàn)了。

在此之前的一周里，我們的股市表現(xiàn)糟糕。道瓊斯工業(yè)平均指數(shù)周五暴跌108.35點，創(chuàng)當(dāng)時歷史上的最差紀(jì)錄。那些在當(dāng)周拿不定主意該怎么做的投資者，在周末下定了決心：如果要打仗了，那就該拋售了。大量投資者在那個周一蜂擁逃離股市，這讓很多人堅定了拋售的決心，以至于大多數(shù)人都忘記襲擊已經(jīng)發(fā)生了。到周一晚上，人們開始擔(dān)憂股市會崩潰，那是我們的股市所遭遇的最慘痛的一天，而在那天上午，人們擔(dān)心的還是戰(zhàn)爭的前景。

現(xiàn)代股市的崩盤，都是由于一些看起來似乎與金融無關(guān)的外部因素所促成的。1906年的舊金山地震導(dǎo)致資產(chǎn)流動性極差，隨著資金涌入西海岸，金融市場逃亡，1907年的金融大恐慌隨之爆發(fā)。1929年，克拉倫斯·哈特立的騙局在倫敦遭到揭露，美國的持股人之后一個月里在紐約大量出售股票。1987年，投資者只用了一個周末來決定拋售。而到2010年5月5日，暴亂、縱火和希臘雅典三名銀行員工遭到謀殺，讓美國的一名資產(chǎn)經(jīng)理在一天之內(nèi)下定決心，采用不顧后果的算法，幾分鐘里賣掉了價值41億美元的巨額股票。接踵而至的閃電暴跌，讓當(dāng)天已經(jīng)下跌300點的道瓊斯工業(yè)平均指數(shù)在隨后13分鐘內(nèi)繼續(xù)狂跌700點，成為了我們所知的最混亂的崩盤。

從一年到一個月，再到一個周末，再到一天，毫不令人意外，催化因素的出現(xiàn)和隨后引發(fā)的崩盤，期間的間隔越來越短，因為進(jìn)行交易、收到確認(rèn)的時間也同樣從幾天變成了幾小時，又變成了幾毫秒。然而這意味著，利用純粹的算法交易系統(tǒng)——它們沒有任何實時的人類監(jiān)控，許多都通過掃描新聞和社交媒體來獲得交易線索——會讓下一次催化事件發(fā)生到引發(fā)崩盤的間隔期只剩一個小時甚至一分鐘。

隨著催化事件和崩盤的間隔期縮短，我們的容錯率也會減少。下一次，我們可能無法做出合理的政策回應(yīng)，再加上只有總統(tǒng)和交易所才有權(quán)提早收盤——即便是證券交易委員會也沒有這個權(quán)力——因此我們甚至都沒時間從決策者那里聽到安慰。

如今的交易已經(jīng)由自動化和算法化主導(dǎo)，因此是時候應(yīng)對這種危險了——不要等到市場崩盤、損害已經(jīng)產(chǎn)生之后，盡管那時我們可能才有勇氣采取行動。我們的各色市場已然經(jīng)歷過小規(guī)模的閃電暴跌，這種趨勢令人不安。除了2010年5月的大規(guī)模暴跌，今年7月的白銀市場、2015年8月的股市、2015年3月的美元和2014年10月的長期國庫債券都出現(xiàn)過崩盤。我們已經(jīng)得到過警告了。

閃電暴跌的情況還會繼續(xù)下去，直到交易所讓那些對此負(fù)有責(zé)任的人承擔(dān)足夠高昂的代價為止。2010年5月6日的閃電暴跌，是一位機(jī)構(gòu)投資人取消了算法中所有關(guān)于價格和時間的防護(hù)措施之后導(dǎo)致的。如果他付出的代價要超過把那些股票低價賣出的損失，那這種情況就不會出現(xiàn)。如果投資者和他們的經(jīng)紀(jì)人之后被拒絕進(jìn)入股市，直到他們獲得了交易所的認(rèn)證，確保缺陷已經(jīng)被修復(fù)，方才解除禁令，那我們或許可以率先阻止閃電暴跌。

任何導(dǎo)致閃電暴跌的經(jīng)紀(jì)人，都應(yīng)當(dāng)接受交易所對源代碼的定期測試，確保其中采用了在追求極致速度時往往被棄用的防護(hù)措施，才能被重新許可進(jìn)入股市。

我們的股市還會崩盤，盡管崩盤這種事看起來還很遙遠(yuǎn)。1907年金融大恐慌和1929年崩盤間隔了二十多年，1987年的黑色星期一和2008年的抵押危機(jī)，期間也有二十多年，然而這樣的事實并不意味著我們可以忽略純電子和算法交易中隱含的缺點。要記住，從抵押危機(jī)到2010年5月的閃電崩盤，期間只有不到兩年。

算法交易者有責(zé)任搞清楚他們的系統(tǒng)會產(chǎn)生怎樣的反應(yīng)。交易所也有義務(wù)保證那些人履行了這一責(zé)任。（財富中文網(wǎng)）

作者斯科特·內(nèi)申斯是《美國五次崩盤的歷史》的作者，也是NationsShares的總裁。

譯者：嚴(yán)匡正

The oil tanker Sea Isle City had originally been named after a rocky island in the Persian Gulf, Umm al-Maradim, which translates, appropriately if unpoetically, into Mother of Boulders. During the second half of June 1987, she and a number of other Kuwaiti ships were rechristened and began sailing with new names and under the American flag in an effort to deter a bellicose Iran from launching threatened attacks on shipping in the Gulf.

It didn’t work and on Friday, Oct. 16, 1987, Iran attacked, launching Silkworm missiles from the Al-Faw Peninsula that struck the Sea Isle City’s bridge and crew quarters. Sixteen were injured and the captain and a lookout were blinded. Two days later the U.S. Navy responded. Operation Nimble Archer destroyed two Iranian oil platforms being used for surveillance and as bases for helicopter and amphibious patrols. Americans waking up on the morning of Monday, Oct. 19, 1987 thought we were finally at war with Iran.

Our stock market had performed terribly the week before; Friday’s loss of 108.35 in the Dow Jones Industrial Average was the worst point loss ever. Those investors who were uncertain about what to do during that week made up their minds over the weekend; if we were at war it was time to sell. The crush of investors fleeing our stock market on that Monday has now overshadowed the catalyst, which convinced many to sell to such a degree that most have forgotten the attacks ever occurred. By Monday evening the worry was that day’s stock market crash, the worst single day our stock market has ever had, but during the first half of the day it was the promise of war.

Every modern stock market crash has been precipitated by some external catalyst that often seems to have nothing to do with finance. But the 1906 San Francisco earthquake led to a crippling lack of liquidity and spawned the Panic of 1907 as capital flooded the West Coast and fled financial markets. In September 1929 the Clarence Hatry fraud was uncovered in London and stockholders in the U.S. decided over the course of the next month to sell in New York. In 1987 it took a weekend to push investors to sell and in 2010 it took just a day from the May 5 rioting, arson, and murder of three bank employees in Athens, Greece to convince one asset manager in the U.S. to launch an unthinking algorithm which sold $4.1 billion worth of stocks, an unthinkable amount, in just a few minutes. The ensuing Flash Crash drove the Dow Jones Industrial Average, which was already down 300 points for the day, down another 700 in just 13 minutes in the most chaotic crash we’ve known.

A year to a month to a weekend to a single day; it should be no surprise that the time from the catalyst to the subsequent crash has collapsed, just as the time it takes to execute a trade and receive a confirmation has similarly collapsed from days to hours to milliseconds. But this means that with the number of purely algorithmic trading systems at work—none of which have any real-time human oversight and many of which are scanning news and social media for trading cues—the time between the next catalyst and start of the ensuing crash may be just one hour or one minute.

As this time span between catalyst and crash has collapsed, so has our margin of error. Next time we won’t be able to formulate a reasonable policy response, and since only the president and the exchanges have the authority to close our markets early—even the Securities and Exchange Commission doesn’t have this power—we may not even have time for reassuring words from policymakers.

With trading now overwhelmingly automated and algorithmic, this is the time to address the danger—not when the market is crashing and not after the damage is done, even though that tends to be when we find the courage to act. Our varied markets have already displayed a troubling tendency to experience smaller flash crashes. Beside the big one in May of 2010, we’ve also seen crashes in the silver market this July, the stock market in August 2015, the U.S. dollar in March 2015, and the Treasury bond market in October 2014. We’ve been warned.

Flash crashes won’t stop until exchanges make them too expensive for those responsible. The May 6, 2010 flash crash was caused by one institutional investor who stripped all the price and time safeguards from their algorithm; they wouldn’t have done that if the cost was more than just selling at disadvantageous prices. If the investor and their broker were both denied access to markets until they could have certified to exchanges that weaknesses had been fixed, we might have stopped that flash crash in the first place.

Any broker who causes a flash crash should only be readmitted to the market after being subjected to periodic testing of their source code by exchanges to assure it is equipped with the sort of safeguards that are often removed in the desperate search for speed.

Our stock market will crash again, although any crash is likely many years off. But the fact that 20 years passed between the Panic of 1907 and the Crash of 1929, as well as between Black Monday 1987 and the mortgage meltdown of 2008, doesn’t mean we can ignore the hidden weaknesses of purely electronic and algorithmic trading. Remember that it was less than two years from the bottom of that mortgage-inspired meltdown to the flash crash in May 2010.

Algorithmic traders have a responsibility to understand how their systems will react. The exchanges have a duty to make certain they fulfill that responsibility.

Scott Nations is the author of A History of the United States in Five Crashes and the president of NationsShares.