算法投資靠譜嗎 AI progress fails to convince all investors

Isaac Newton may have been one of the finest minds of all time, but he turned out to be a miserable investor. “I can calculate the motions of the heavenly bodies, but not the madness of people,” he lamented after losing a fortune in the South Sea bubble.

艾薩克•牛頓(Isaac Newton)可能是有史以來最聰明的人，然而事實證明他是個糟糕的投資者。“我能計算出天體的運動，卻不能計算出人的瘋狂，”他在南海股票泡沫中損失了一大筆錢以後哀嘆道。

Increasingly, however, technology-savvy investors think they can harness mathematics and bleeding edge computer science to predict the ebb and flow of financial markets. Some of the most advanced asset managers are turning to artificial intelligence techniques, with investment algorithms that can autonomously learn, adapt and scour vast data sets for tradable patterns.

然而，越來越多精通技術的投資者認爲他們可以利用數學和計算機尖端科技，來預測金融市場的起起伏伏。一些最先進的資產管理公司現在正求助於人工智能(AI)技術，其中包括能夠自動學習、適應和搜索大量數據組以研究出可交易的模式的投資算法。

But some “quantitative” financiers (quants) are sceptical that these tools are any more than a somewhat better mousetrap, and argue that areas such as “machine learning” are overhyped and AI used as a marketing gimmick.

但有些“量化”金融家(quant，即量化分析師)懷疑這類工具可能不過是一種高明一點的陷阱。他們認爲“機器學習”這類領域被過度炒作，AI則是一種營銷噱頭。

“Everyone wants the Holy Grail, something they can invest in and it will make 1 per cent a month forever,” says Ewan Kirk, head of Cantab Capital, a Cambridge-based quantitative hedge fund. “I don’t want to be cynical, but I am sceptical.”

“每個人都想要得到‘聖盃’，某種能夠投資並且實現1%恆定月回報率的東西，”位於劍橋(Cambridge)的量化對衝基金Cantab Capital的負責人尤安•柯克(Ewan Kirk)表示，“我不想表現得悲觀，但我很懷疑。”

David Harding, head of Winton Capital, one of the biggest quantitative hedge funds in the world, is also doubtful that AI represents a quantum leap for the investment industry. “I’m not a Luddite, we’re always interested in new ways to make money. But I have to be very sceptical because I constantly have world-class people showing me miracle cures that don’t actually work,” he says.

全球最大量化對衝基金之一溫頓資本(Winton Capital)負責人戴維•哈丁(David Harding)也懷疑，AI並不能給投資業帶來重大飛躍。“我不是盧德分子(Luddite)，我們總是對賺錢的新方式感興趣。因爲總有世界級的人物向我展示實際上並沒有效果的靈丹妙藥，我不得不對此深表懷疑，”他說。

Dramatic improvements in computing power have revolutionised the investment world, with algorithmic traders and investors increasingly influential across markets. Money is pouring into computer-driven hedge funds that have consistently managed to parse signals amid market noise. As a result many money managers are scrambling to hire computer scientists, often pitting them in direct competition for talent with Silicon Valley’s tech giants and hot start-ups.

計算能力的顯著提升徹底改變了投資界，依據算法的交易商和投資者在市場上的影響力越來越大。大量資金涌入持續從市場雜音中分析出風向的計算機驅動對衝基金。這導致許多資金管理公司競相僱傭計算機專家，直接與硅谷技術巨頭和熱門初創企業爭奪人才。

AI is at the forefront of this. The field has also enjoyed several leaps forward in recent years. Most notably, Google’s DeepMind AI arm has created a programme that recently thrashed a legendary player of Go, an ancient Chinese game that is so complex that most experts previously reckoned it would take at least a decade before a computer could beat a human champion.

AI處於領域的最前沿。近年來AI領域也經歷了幾次飛躍。最引人注目的是，谷歌(Google)旗下DeepMind的AI部門研發的程序，最近打敗了一位著名圍棋選手。圍棋是一種古老的中國遊戲，因爲過於複雜，大多數專家此前都認爲，計算機至少還需要10年才能打敗人類圍棋冠軍。

The potentially wider applications of techniques used by the likes of DeepMind’s AlphaGo algorithm has fuelled optimism that investment management could be on the cusp of another technological revolution, possibly similar in scale to the electronification of markets in the 1970s and 1980s.

DeepMind的AlphaGo這類算法所運用的技術或許還能得到更廣泛的應用，這引發了有關投資管理可能即將迎來另一場技術革命的樂觀情緒。在規模上，這場革命可能和上世紀七八十年代的市場電子化革命相仿。

“Machine learning and artificial intelligence is going to play a very large role in quant managers, but also with traditional asset managers that are aggressively expanding in this space,” says Osman Ali, a fund manager at Goldman Sachs Asset Management.

“機器學習和人工智能將在量化資產管理中起到極大作用，但傳統資產管理公司也會在這個領域大舉擴張，”高盛(Goldman Sachs)資產管理部門的基金經理奧斯曼•阿里(Osman Ali)表示。

Popular AI approaches such as machine learning can be used by computers to learn and develop autonomously. For example, a machine learning algorithm can learn to play and master a computer game such as Super Mario independently, at first playing the arcade classic randomly but quickly figuring out how the controls work and how to get to the end of the level.

計算機可以利用機器學習等流行的AI策略自主學習和發展。比如，一種機器學習算法可以獨立上手和掌握如何玩《超級馬里奧》(Super Mario)這樣的遊戲。一開始算法會隨機地玩這款經典街機遊戲，但很快算法就能摸清如何操作和通關。

There is therefore widespread enthusiasm over the potential of unleashing machine learning algos to find fleeting but profitable patterns in the vast sea of data.

因此，自由的機器學習算法在海量數據中尋找稍縱即逝的可盈利模式的潛能，引起人們的廣泛興趣。

“I think of algos as little children that can scale tremendously. And you can teach them to read millions of books at the same time,” says Brad Betts, a former Nasa computer scientist working in BlackRock’s San Francisco-based Scientific Active Equity arm.

“我認爲算法就相當於擁有巨大潛力的幼童。你可以教它們同時閱讀數百萬本書，”美國國家航空航天局(NASA)前計算機科學家、現在供職於貝萊德(BlackRock)位於舊金山的“科學主動股票投資”部門的布拉德•貝茨(Brad Betts)表示。

Yet scepticism, even among many quants, is still pervasive. They see areas such as machine learning and deep learning — the latter underpinned DeepMind’s Go exploits — merely as extensions or enhancements of techniques that have for long been in use.

然而，甚至是在很多量化分析師中，懷疑情緒依然普遍。在他們看來，機器學習和深度學習——後者支撐了DeepMind的AlphaGo引人注目的成功——只不過是對已經投入使用很長時間的技術的擴展或加強。

“Lots of people use techniques that could be called machine learning for decades,” argues Robert Hillman, head of Neuron Capital. “There’s a huge difference between image recognition and using AI in markets. Will this be a paradigm change for investing? I don’t think so … It’s not a fundamental change, it’s an efficiency improvement.”

“很多人使用了數十年的一些技術，都可以被稱爲機器學習技術，”Neuron Capital負責人羅伯特•希爾曼(Robert Hillman)表示，“圖片識別和把AI運用到市場之中存在巨大差異。這是否將帶來投資的範式轉變？我不這麼認爲……這不是根本性的變化，這是一種效率的提升。”

Mr Kirk points out that most common AI approaches are focused on pattern recognition, such as telling the difference between a cat and a dog in an image. But markets are dominated by noise and chaos, the patterns are harder to find.

柯克指出，最常見的AI策略着重於模式識別，比如區分出圖片中的一隻貓和一隻狗。但市場上充斥着雜音和亂流，要找到模式更爲困難。

“As a geek I’m super-excited about AlphaGo, but it’s a big leap from beating a game with clearly defined rules and objectives and investing,” he says.

“作爲一名極客，AlphaGo讓我超級興奮，但從打贏一個有清晰規則和目標的遊戲、到進行投資，中間還有巨大的跨度，”他說。

Even quants that are cautiously optimistic on the future of AI in investing warn of many pitfalls. Algorithms that may look ingenious and backrest superbly against historical data have a nasty habit of unravelling when confronted with unforgivingly fickle financial markets.

即使是對AI在投資界的應用前景抱謹慎樂觀態度的量化分析師，也警告這個領域存在許多陷阱。一些看起來可能很巧妙、與歷史數據完美契合的算法，在面對金融市場的反覆無常時卻常常出毛病。

“Playing Super Mario might not necessarily work for markets. If you hit the button you always know what will happen, but you don’t in markets,” says another quant at a large hedge fund. “It can take time for it to find the good trades and to optimise them. It can go through a lot of bad trades.”

“能玩《超級馬里奧》未必能駕馭市場。當你按下按鍵的時候，你總是知道會發生什麼，但在市場上就不是這樣了。”一家大型對衝基金中的另一名量化分析師表示，“算法可能需要時間才能找到好的交易機會並進行優化，可能先要經歷很多糟糕的交易。”