交易層面的先進量化與落後量化區別原則

作者：Ralph Cheng 鄭宇和

自從去年成立了自己的量化基金，我發現多數人對於量化金融並不怎麼瞭解，遑論如何辨別先進量化與落後量化的區隔。事實上，不僅僅是一般大眾，甚至連業界專業人士有時也很難明白量化金融究竟是什麼。更糟的是，有些專業人士甚至假裝明白。

不過這種情形也不是不能理解，畢竟量化一詞本就包羅萬象，容易引發混淆。舉例來說，各種不同的交易策略、定價公式，或風險管理方法，只要涉及了複雜的數學計算，都有可能被納入量化的領域。這篇專文則將針對交易層面的量化金融，闡釋幾個基本概念。

初階應用：把技術分析寫進自動化交易策略

就最初階的應用而言，許多散戶交易員會將技術性分析的進場與退場規則，寫進自動化的交易策略程式裡。例如，將某一金融工具價格的短期與長期移動平均值的黃金交叉，設定為自動執行買進的訊號。這種情況下，金融工具的價格，乃是唯一的一筆輸入資料，於是乎問題就非常簡單了：我們能否僅僅參照過往歷史價格，就完成未來價格的量化作業？

這樣當然不夠，金融市場的實際複雜程度，遠遠超過這種方法所能因應的範圍。儘管如此，只要能運用得當，這麼做還是在一定程度上的降低風險、提高效率。

中階應用：交易策略涵蓋多項因子

從中等階層的應用開始情況越趨複雜。這個階段所建構的交易策略，所參考的因子不只一個。例如在美國股票交易策略上，同時必須將各種類別的美國經濟數據以及公司的財務報告數據放入考量。當然，這必須具備財金及經濟學上一定程度的知識，才有辦法做到。

這個階段裡，藉由足夠充分的研究考察，交易員得以辨識出哪些投資策略得以歷經時間的考驗而締造獲利。縱然如此，這個階段仍然不保證投資策略的報酬必然十拿九穩，許多未被納入模組的市場事件與價格起伏仍可能造成巨幅貶損。

高階應用：包含人工智慧及高頻交易

而在專業投資的領域裡，量化則不再拘泥於線性函數，而是進一步邁向更高維度的數學。專業的量化金融所應用的策略中包含了人工智慧以及高頻交易。現今人工智慧領域中的機器學習常用於處理人腦無法察覺的模式與邏輯序列，在「大數據」當中掌握各種可用線索。

如同「量化」一詞，「大數據」一詞也經常被誤用，多數人並未領略大數據的真實意涵。簡而言之，大數據乃是無法以傳統統計方法處理的龐雜資料。針對大數據加以解讀之後，並不是說所有造成策略虧損或獲利的因子與邏輯都被包含，不過因此產生的模型相較於舊有模型在解釋市場上確實較為優良。然而，人工智慧策略的建構並不容易，同時仰賴了電腦科學、統計、數學及財務金融等高階技能。

至於高頻交易策略，則是依賴高速送出買賣訊號，希冀藉由小幅價差反覆累積獲利。這是以奈秒 （nanoseconds）計算的超迅速戰爭，涉及層面含括了軟體及硬體。勝負取決於處理器速度、程式語言的選擇、程式的效率、連線速度，甚至連伺服器的所在位置都有顯著影響。最終則演變成為高頻交易公司間的軍備競賽。它的進入門檻極高，並不是每個市場都適合或允許運用此種交易策略。

以上僅是粗淺的說明，針對博大精深的相關領域稍做釐清。我認為量化一詞確實是個非常廣泛的概念，本文所稱者則僅限於交易層面的探討。不過，即便只是基本概念層次的量化交易簡介，仍可協助讀者對做量化交易的朋友們提出切中要點的質問，並從中探出其對量化交易上的深度與了解。

更多文章請見法意專欄好文

A Framework for Distinguishing Advanced and Backward Quants in Trading

Since starting my own quant fund last year, I have run into many people who have little idea about quant, let alone how to distinguish more advanced quants from backward ones. Indeed, sometimes not only the public but also financial professionals have a hard time knowing what a quant is—and, even worse, some professionals will pretend that they do know. But all this is entirely understandable, because quant is a very broad term that can often be a source of confusion. For example, it can represent many different trading strategies, pricing formulas, or approaches to risk management, all of which involve complicated math. In this article, I will convey some basic ideas having to do with quant in trading.

At the beginner level, we find many retail traders coding technical analytic entry and exit rules into automated trading strategies. For example, one might program for a financial instrument’s price a short-term moving average crossing above the long-term into an automatically executed buy signal. At this level, an instrument’s price is the sole input datum, and this approach essentially comes down to a question of whether a financial instrument’s future prices can be quantified merely with reference to its past prices. Of course, they cannot; the financial market is much more complex than such an approach would imply. However, this approach is still, to some extent, a useful tool for helping traders reduce risk and enhance their work efficiency—if used properly.

At the medium level, things get a little bit more complex. Here people start begin building trading strategies that rely on more than one factor. For example, one might base a U.S. equity trading strategy on all kinds of U.S. economic data as well as on companies’ financial report data. Certainly, doing so would require a certain degree of knowledge about finance and economics. However, at this level, with enough research and study, a trader can identify strategies that produce profits over time. Even so, at this stage, the stability of a strategy’s return still cannot be guaranteed, for events and price movements not included in the model can create huge drawdowns.

In the professional arena, we move from linear to higher dimensions. Here professional quants use strategies that rely on artificial intelligence and high-frequency trading. Current applications for machine learning in the field of artificial intelligence allow traders to deal with “big data” that contains patterns and logical sequences not yet noticed by humans.

Just like the term quant, the term big data is often misused by the public, which has little understanding of what big data really is. Generally speaking, big data is data that cannot be processed using traditional statistic methods. After unlocking big data, traders might not have included all factors and logic needed to avoid all drawdown, but the resulting model does have decided advantages over its predecessors. However, building an AI strategy requires high-level skills in computer science, statistics, math, and finance.

High-frequency trading, by contrast, is a strategy that relies on sending buy and sell signals at extremely high speeds in hopes of profiting from small spreads. This war is fought in nanoseconds and involves not only software but also hardware. Its battles are won by processor speed, choice of programming language, efficiency of code, connection speed, and even server location. At its apex, it becomes an arms race between companies that specialize in high-frequency trading. The barrier to entry is extremely high—and not every market is even allowed or suitable for such a trading strategy.

Even these explanations only scratch the surface of what remains to be learned in this sphere. I have noted that quant is a very broad word, and this article, though touching on aspects of its meaning, has restricted itself to the field of trading. However, even these basic conceptions of the role of quant in trading will allow the reader to get some idea of the skill of his or her quant friends by asking informed questions about how they build their strategies.

★延伸閱讀★
計量金融分析師如何看待升息這件事？
法國大選中的國家陣線
物理學家走入華爾街 擦出什麼火花？
機器人該繳稅嗎？

【作者簡介】Ralph Cheng 鄭宇和

海外量化對沖基金經理人
畢業於西雅圖華盛頓大學 University of Washington – Seattle
Acquisition International 2017 東亞最具創新基金經理人
Acquisition International 2017 年度最佳對沖基金