The lesson from the most recent quant quake
最近一次“量化地震”的教训

机器卡壳 Sand in the gears-书迷号 shumihao.com

WHAT IS IT like to lose to a machine? In 1997 the world’s best chess player, Garry Kasparov, was beaten by Deep Blue, a $10m super-computer made by IBM. Twenty years later he wrote “Deep Thinking”, a book about the experience. What comes across vividly is how exhausting each game was. Chess players, even great ones like Mr Kasparov, get tired and frustrated. Doubts begin to creep in. By contrast, Deep Blue just needed the occasional reboot.
输给一台机器是什么感觉?1997年,世界上最优秀的国际象棋选手加里·卡斯帕罗夫(Garry Kasparov)被一台价值千万美元的IBM超级计算机“深蓝”打败。20年后,他围绕这段经历写下了《深思》(Deep Thinking)一书。其中让人感受最深的是,每一场比赛都如此让人精疲力竭。即使像卡斯帕罗夫这样伟大的棋手也会感到疲倦和沮丧。自我怀疑随之滋生。相比之下,深蓝只需要偶尔重启一次。

Now turn the tables. What is it like to win against the machines? By New Year’s Eve the least smart buy-and-hold investor in an index fund might be able to boast of such a victory. For 2020 has been rotten for “quant” funds, which use powerful computers to sift market data for patterns that might predict future prices. “Long-short” momentum—buying recent winners and selling recent losers—had been one of quant’s better strategies this year. Yet on November 9th, when news broke of an effective vaccine for covid-19, it had its worst ever day.
现在换位想一下。击败了机器又是什么感觉?到今年年底,哪怕是最不精明的投资者,只要买入并持有一支指数基金,就可能取得这样的胜利。因为对于使用强大的计算机从市场数据中寻找模式来预测未来价格的“量化”基金来说,2020年是惨烈的一年。买入近期表现好的股票、卖出近期表现差的股票的“多空”动量策略曾是今年表现较好的量化投资策略之一。然而,11月9日有关一种新冠肺炎疫苗有效的消息传出时,该策略经历了有史以来最糟糕的一天。

Quants rely on history. If something happens that is without precedent, such as a vaccine in a pandemic, they have a problem. No doubt a few quant hedge funds are nursing heavy losses. And perhaps a few discretionary funds have made a killing. The terrain on which human traders can beat the machines is much diminished. But November 9th shows it is still possible. Chalk it up as a small victory for the species.
量化交易依赖历史数据。如果发生的事件没有先例,例如在全球疫情中一种疫苗问世,它们就有麻烦了。几个量化对冲基金无疑在经受巨亏。而一些自选型基金或许大赚了一笔。人类交易员能够击败机器的领域已经不多了,但11月9日表明这仍然是可能的。就当这是人类取得的一次小小胜利吧。

It is no small irony that momentum trading takes advantage of human weaknesses. One of these is “conservatism bias”. Investors tend to stick to prior views too rigidly and change them only slowly in response to new information. They may give undue emphasis to the price paid for a stock as a marker of its true value and, as a consequence, sell winning stocks too soon and hang on to dud stocks for too long. There is also a contrasting tendency to extrapolate past success. So as well as under-reacting to news, people also over-react to it. Momentum trading seeks to exploit this.
颇为讽刺的是,动量交易策略恰恰是利用了人类的弱点。其一是“保守性偏差”。投资者往往过分坚持自己先前的观点,面对新信息时不能及时改变看法。他们可能会过分强调用股票的买入价格来衡量其真实价值,因此会过早地卖掉牛股,而又过久地持有表现平庸的股票。另一种相反的倾向是根据过往的成功推断未来。因此除了对新消息反应迟缓之外,人们也会反应过度。动量交易正是试图利用这一点。

A lot of long-short strategies, including momentum, rank stocks by a particular attribute and then buy the top decile (or quintile) of the group and sell the bottom one. This requires machines. Sorting through thousands of securities quickly is beyond the meagre talents of a living, breathing portfolio manager. It requires algorithms that first establish and then fine-tune the optimal period over which to do the sorting. And it needs speedy and seamless access to automatic trading platforms and market data. You would not want to do all this by hand and brain.
包括动量交易在内的许多“多空”策略都是根据特定的属性为股票排序,然后买入该类别当中的前十分位(或五分位),同时卖出垫底的股票。这种操作需要机器完成。单凭人类投资组合经理的薄弱能力无法迅速筛选成千上万的证券。这需要通过算法来确定并微调出一个最优的筛选期。同时也需要快速、无缝地访问自动交易平台和市场数据。这样的工作是人的双手和大脑无法胜任的。

In chess, the brute force of computing power eventually wins out. In investing, the strength of synthetic traders is in dealing with reams of information that is machine-readable, such as tick-by-tick stock prices. The most powerful machines can make sense even of unstructured (“big”) data. But an event like the discovery of a vaccine can flummox even the smartest of them. Humans retain an edge. They are able to winnow down endless possibilities using mental shortcuts. They can imagine scenarios that the past has not thrown up—scenarios such as “a vaccine may become available soon, given the amount of money and effort being thrown at it”; and “news of such a vaccine might spark a sell-off in ‘stay-at-home’ shares and a rally in ‘get-out-of-the-house’ shares”.
在国际象棋中,机器运算的蛮力最终胜出。在投资中,机器交易的优势是能够处理大量机器可读的信息,例如一笔笔的股票交易价格。最强大的机器甚至还能读懂非结构化数据(“大数据”)。但是,即便最聪明的机器面对疫苗诞生这样的事件也会不知所措。人类仍有一个优势。他们能够利用心理捷径来筛选无限的可能性。他们可以想象过去从未出现过的情景——比如,“都已经投入这么多钱、付出这么多努力了,一种疫苗也许很快就会面世”;以及“有关这种疫苗的消息可能会引发‘居家’股票的抛售和‘出门’股票的反弹”。

But why were the moves in prices so dramatic? A good rule of thumb, says one quant guru, is that the faster you trade, the less capacity there is for your strategy. A speedy trading strategy, such as momentum, relies on liquid markets to keep turnover costs in check. The strategy can become crowded. And when the quants suffer losses, they may be forced by risk-management rules to close their positions. As everyone rushes to get out at the same time, it makes for extreme price movements. This is in part why sophisticated quant funds are constantly evolving. They look for unique datasets on which to train their machines. Or they try to come up with new ways to parse weaker signals that others cannot detect in the market noise.
但为何价格波动如此剧烈?一位量化交易专家表示,一条很好的经验法则是,一种策略的交易速度越快,可供操作的市场容量就越小。像动量交易这样的快速交易策略需要有高流动性的市场来控制交易成本。可能会有太多交易采用这种策略。一旦量化基金蒙受损失,可能会根据风险管理规则的要求被迫平仓。由于所有人都在同一时间争相离场,就会导致价格的极端波动。这也是先进的量化基金不断演化的原因之一。它们寻求独特的数据集来训练机器,或者试图采用新的解析方法从市场噪音中辨别出别人无法察觉的微弱信号。

The quants have had a rough time, but they are hardly in retreat. Their domain will only expand. The margin of advantage for discretionary trading—for human ingenuity, in other words—will shrink. It is worth remembering that the first time Mr Kasparov played against Deep Blue, in 1996, he won. Now, as he has pointed out, you can download free chess engines that are far more powerful. We should savour victories over the machines while we can.
量化交易这次栽了跟头,但丝毫没有退让。它们的领地只会越来越大。自选交易——也就是人类的聪明才智——相对于机器的优势将逐渐缩小。值得一提的是,1996年卡斯帕罗夫第一次与深蓝交手时,他赢了。而现在,正如他指出的,你能免费下载的国际象棋程序都比当年的深蓝强大多了。趁我们还能击败机器的时候,细细品味这胜利的滋味吧。