對量化投資的批評

對量化投資的批評

 

筆者為了研究適用於強積金的投資策略,遍尋各門各派,從經典名著至坊間blog文,從諾貝爾獎大師到市井小民,筆者都嘗試了解其理論及操作,然後放於強積金環境測試回報波幅及跟蹤觀察,亦有考慮其他因素如可靠性、簡易性、長久性等等,將較好的7種策略收入《積金大反擊》一書。其中第5至第7招偏向運用量化(Quantitative)方式,較為機械化操作。

 

量化投資於近10至20年外國的投資界大行其道,尤其對沖基金都大量依賴量化方式進行投資。期間有不少人對量化投資提出質疑,下面主要引述近期於Bloomberg網站的一篇評論文章進行討論:

 

 

評論文章原文 中文版(Google Translate) 筆者觀點
Like alchemists seeking to transform base metals into gold, asset managers are constantly seeking the perfect formula for investment success. In recent years, the search has focused on proprietary quantitative strategies, involving rule-based investments. The genre is ill-defined and marketed under different names, including factor investing, risk parity, smart beta and so forth.

 

與尋求將基本金屬轉化為黃金的煉金術士一樣,資產管理公司也在不斷尋求投資成功的完美方案。近年來,搜索專注於專有定量策略,涉及基於規則的投資。該類型的定義不明確,並以不同的名稱進行營銷,包括要素投資,風險平價,智能beta等。 見前文 1  2  3
Whatever the rubric, “quant” funds now have over $1.5 trillion under management. Index and quantitative investing account for over half of all equity trading, double the level a decade ago. But, after initial success, they’ve produced uneven returns recently — a bad run that may well continue. Arguably, the inherent weaknesses of the approach are now being exposed.

 

無論什麼標題,“量子”基金現在管理著超過1.5萬億美元。所有股票交易中,指數和定量投資佔一半以上,是十年前的兩倍。但是,在最初的成功之後,他們最近產生了不均衡的回報 – 這可能會繼續下去。可以說,這種方法的固有弱點現在正在暴露。 因為有優點才多人用。
Fancy math can’t mask the fact that quantitative investing strategies essentially rely on pattern recognition: Models look to correlate past periods of superior returns with specific factors including value, size, volatility, yield, quality and momentum. Once the latter are identified, fund managers construct portfolios with specified return and risk parameters consisting of securities that match those optimal characteristics. Other techniques exploit short-term dislocations between individual prices and comparable securities or the broader market, betting that the relationship will eventually revert to normal.

 

花哨的數學不能掩蓋這樣一個事實:定量投資策略主要依賴於模式識別:模型看起來將過去的優質回報期與特定因素相關聯,包括價值,規模,波動性,收益率,質量和動量。一旦識別出後者,基金經理就會構建具有特定回報的投資組合,並且風險參數由與這些最佳特徵匹配的證券組成。其他技術利用個別價格與可比證券或大盤之間的短期錯位,認為這種關係最終將恢復正常。 這個是量化投資理論的缺憾,只有觀察整合過往的經驗及數據,運用induction方式去獲取知識,是本質上的不全面;但經濟及投資等人文科學,其本質便是如此不確定,不像自然科學般可以用deduction去獲取知識。
Such approaches have several fundamental weaknesses. First of all, quant investing is tainted by hindsight bias — the belief that understanding the past allows the future to be predicted. Given enough time, money and computing power, a strategy predicting high returns can be found and validated using back testing to check its historical performance. But, this heightens the risk of overfitting, or adjusting the model to suit a specific set of historical conditions. Those may look like a winning recipe, but could turn out to be an historical fluke.

 

這些方法有幾個根本的缺點。首先,量化投資受到後見之明的偏見的影響 – 相信理解過去可以預測未來。如果有足夠的時間,金錢和計算能力,可以使用回溯測試找到並驗證預測高回報的策略,以檢查其歷史性能。但是,這會增加過度擬合的風險,或者調整模型以適應特定的歷史條件。這些可能看起來像一個勝利的食譜,但可能會成為一個歷史性的僥倖。 Over-fitting是一般量化投資入門者的通病,因為優化結果太吸引;進階者反而處處提防優化的影響,例如《積金大反擊》用多計劃、多時間間距,用分散投資(diversification)去減少對單一系列參數的依賴,並定期檢討反省。
Modelling is also affected by practical matters, such as what data is or isn’t available. London Business School researchers found over 300 factors that could be used to develop potential strategies, heightening the risk of an overfitted model. There is in addition the problem of ergodicity, that is, the lack of a truly representative data sample.

 

建模也受實際問題的影響,例如哪些數據是可用的或哪些不可用。倫敦商學院的研究人員發現了300多個可用於製定潛在策略的因素,從而提高了過度擬合模型的風險。此外還存在遍歷性問題,即缺乏真正具有代表性的數據樣本。 現時學界已有非常大量的研究及數據,不存在對數據有篩選而「作數」。
It’s important to remember, too, that financial eras are characterized by specific policies, market structures, instruments and investors. Unique conditions that shape returns, volatility and correlation may change. While models create an illusion of sophisticated certainty, they can’t capture the full range of events that produced a particular outcome and could perform poorly where a paradigm shift occurs. Modern markets may simply be too complex to be modeled accurately.

 

重要的是要記住,金融時代的特點是具體政策,市場結構,工具和投資者。形成回報,波動性和相關性的獨特條件可能會發生變化。雖然模型創造了複雜確定性的幻覺,但它們無法捕捉產生特定結果的所有事件,並且在範式轉換發生時可能表現不佳。現代市場可能過於復雜而無法準確建模。 政策、工具和投資者的確影響市場,主要視乎「人」的投資行為及心理有沒有變化。

 

任何投資策略都會受政策、工具和投資者影響,不單只量化投資。

 

而且A.I.的發展更能處理人腦所處理不到的複雜。

Quant strategies naturally lack transparency, given that asset managers are reluctant to disclose too many details and lose their competitive edge. This, however, increases the risk of gaming. A low-volatility fund, for instance, might buy illiquid assets whose prices change infrequently, thus giving the illusion of stability. Some strategies, such as selling options, might produce a lot of small gains but be vulnerable to large losses if market conditions change.

 

由於資產管理者不願透露太多細節並失去競爭優勢,因此定量策略自然缺乏透明度。然而,這增加了遊戲的風險。例如,低波動性基金可能會購買價格不經常變化的非流動資產,從而產生穩定的假象。一些策略,例如賣出期權,可能會產生很多小幅收益,但如果市場情況發生變化,則容易遭受巨額虧損。 搵食者有不傳之秘是必然的事,正如可口可樂有秘方一樣。

 

賣出期權策略失敗(見前文)是人的問題,與量化投資沒有必然的因果關係。甚至乎量化投資的機械式操作更能撇除人為偏差。

Increases in funds under management may create competition that reduces a strategy’s expected return. Where the size of funds increases, the strategy could become crowded, making trading difficult and creating unpredictable profits and losses.

 

 

管理資金的增加可能會產生競爭,從而降低戰略的預期回報。在資金規模增加的情況下,戰略可能變得擁擠,使交易變得困難,並產生不可預測的利潤和損失。 Overcrowding這個論點有很多人討論過,結論是量化投資現時對比整個投資市場仍然是少數,還未到達自我毀滅的critical mass。

 

對比外國,量化投資於香港只是初創起步,連一隻像樣的量化投資基金也未有,更遑論強積金市場。

 

At their most basic level, quant funds are selling hope with a financially catchy name. Investors disillusioned with below-expected returns are switching to low-cost index products. To counter this trend, fund managers have sought to seduce investors with complex and opaque black-box strategies, made credible by the rocket science on which they’re supposedly based. Fund managers can then collect their high fees (the standard 2 percent on assets under management plus an additional 20 percent of outperformance) on the promise of future performance.

 

在最基本的層面上,定量基金正在以一種經濟上吸引人的名字賣出希望。投資者對低於預期的回報感到失望,正轉向低成本指數產品。為了應對這一趨勢,基金經理們試圖通過複雜和不透明的黑盒策略來引誘投資者,這些策略因其所謂的火箭科學而變得可信。然後,基金經理可以收取高額費用(管理資產的標準2%,另外還有20%的表現),以表示對未來業績的承諾。 Burton Malkiel於漫步華爾街(A Random Walk Down Wall Street)的2016年新版本中,亦有批評Smart Beta的回報沒有高於大市,而且量化投資換手率高,交易成本蠶食了回報。對沖基金的管理費的確是高,但量化投資亦有ETF,管理費算是低。

 

強積金沒有換手交易成本,股票類成份基金的管理費一般為1%至1.5%。

While there will always be some standouts, it’s not clear why so many managers can claim sustained superior performance. The basic technology, data and expertise is readily available. Logically, the anomalies that the strategies rely on should dissipate. There is an inherent contradiction in that the approach exploits inefficiencies, but requires market efficiency to realign prices to generate returns.

 

雖然總會有一些突出,但不清楚為什麼這麼多經理人可以聲稱持續優越的表現。基本的技術,數據和專業知識隨時可用。從邏輯上講,戰略所依賴的異常應該消失。存在一種固有的矛盾,即該方法利用低效率,但需要市場效率來重新調整價格以產生回報。 量化投資是一個理論,而不只是個別投資經理的秘技,一般人亦可按理論而自己進行操作。

 

由於Overcrowding還未成形,所以異常是不會消失。而量化投資正正是捕捉效率市場內出現的異常(anomaly)

The reality is that any fund managers possessing a magic investment formula guaranteeing low risk and high returns would have no incentive to share the secret. Successful firms such as Renaissance Technologies LLC have closed some funds to outside investors, preferring to capture the returns for themselves. As legendary investor Paul Tudor Jones once noted, if there was a single easy formula to follow, then all investors would already be rich.

 

現實情況是,任何擁有保證低風險和高回報的神奇投資公式的基金經理都沒有動力去分享這個秘密。像Renaissance Technologies LLC這樣成功的公司已經向外部投資者關閉了一些資金,他們更傾向於為自己獲取回報。正如傳奇投資人保羅·都鐸·瓊斯曾經指出的那樣,如果只有一個簡單的公式可供遵循,那麼所有投資者都已經富裕了。 AQR的Cliff AsnessBridgewater 的Ray DalioResearch Affiliates 的Rob Arnott是量化投資的典型分享者。

 

一些基金封盤不接受新投資者,一般是其策略於適用的市場到了頂峰,基金經理又已經名成利就,無需用AUM來證明自己。基金封盤與公不公開投資策略及持股無關(見前文的13F

 

畢菲特的巴郡都沒有發新股集資,可以算關閉了新增投資者,但畢菲特的投資方式簡單可循,亦出現明顯大量真實(及自我幻想)的富裕遵循者。

 

 

正反意見都要檢討反省,然後擇善固執。筆者無期待單憑量化投資可以過世,但投資策略要不斷改良,並加入更加新穎的投資理論及策略(例如8、9、10招)。暫時來說,量化投資於強積金的表現仍是眾多投資理論及策略之中最好與及最適用。