【重磅】年度寬客 (2000 - 2019)
原標題:【重磅】年度寬客 (2000 - 2019)
作者:王聖元
時間就這樣悄無聲息的溜了
2018年,就只剩下26天了
本文獻給剛剛出生的兒子
前幾天看到了 Quant of the Year 2019 頒布的新聞,回想從 2015 年開始自學機器學習時就沒關注這個了,因為這個獎項通常都是 Q-quant (即在風險中性測度下玩轉的 quant) 所拿,而我開始對機器學習感興趣,已經向 P-quant 靠攏了。關於 P-quant 和 Q-quant 的區別可參照我之前寫的這個帖子。但今天這個獎項是頒給一個渣打銀行 (SCB)的數據分析執行董事,他用 P-quant 的機器學習方法來解決銀行實際的問題。心血來潮把從 2000 年到 2019 年的這 20 年的新聞讀了,論文也從 Risk 網站上下載了。(Risk 網站下載這些論文需註冊會員,因此有些論文上有水印不讓傳播,為了尊重知識產權,只在公眾號後台分享那些沒打水印的論文,望理解)
文末下載
論文在文末下載
下面就簡要回顧下這 20年年度最佳寬客做了什麼吧。
2019 年
Alexei Kondratyev
論文 1:Curve Dynamics with Artificial Neural Networks
論文 2:Evolutionary Algos for Optimising MVA
Quants these days tend to maintain expertise in specific fields. With Alexei, his expertise in multiple, unrelated fields gives him a broader perspective and makes him a great researcher. -- Alexander Sokol
新聞鏈接:
https://www.risk.net/awards/6159246/quant-of-the-year-alexei-kondratyev
亮點:用機器學習的法來解決賣方 (sell side) 的問題。
機器學習其實在金融上的應用主要都在買方 (buy side), 比如私募或者基金,而 Alexei 的主要貢獻是在賣方如銀行中找到了兩個應用:
- 用人工神經網路 + 正則化,和自動編碼器來捕捉利率曲線和商品遠期曲線里的動態關係。【論文 1】
- 用遺傳演算法 (Genetic Algorithm, GA) 和粒子群優化 (Particle Swarm Optimization, PSO) 來為銀行壓縮交易而減少保證金 (margin)。兩個都是進化演算法,GA 主要在離散型變數空間 (比如貨幣 currency, 交易對手 counterparty) 找最優解,而 PSO 主要在連續性變數空間 (比如年限 tenor, 本金 notional) 找最優解。【論文 2】
論文還要好好讀,至少現在我覺得第一篇的 input 的選擇就有些不合理,可能犯了數據窺探 (data snooping)的錯誤。Alexei 目前還在研究量子計算 (Quantum Computing),和 NASA 合作把量子計算應用在一個含 60 個資產的組合優化上,節省了一半的計算時間。(是不是有點小題大做了?)
2018 年
Leif Andersen
Michael Pykhtin
Alexander Sokol
論文:Does Initial Margin Eliminate Counterparty Risk
They looked at the entire complexities of the margining process and modelled it mathematically. They looked at things from first principles and the result was amazing.-- Alexei Kondratyev
新聞鏈接:
https://www.risk.net/awards/5371021/quants-of-the-year-leif-andersen-michael-pykhtin-and-alexander-sokol
亮點:深挖交易對手違約後的細節,將保證金風險期 (Margin Period of Risk, MPOR) 分解成四個時段來分析並量化了之前從來沒有人去想要量化的結算風險 (settlement risk)。
2017 年
Jean-Philippe Bouchaud
論文:Cleaning Correlation Matrices
It is really more of a physics approach, to let the data speak. A lot of models used in mathematical finance seem to be more driven by their convenience and the possibility to answer a question with a number, rather than taking the time and thinking about the problem. -- Bouchaud
新聞鏈接:
https://www.risk.net/risk-magazine/analysis/2479713/quant-of-the-year-jean-philippe-bouchaud
亮點:用實證研究 (emprical reseach) 和數據,而不是用理論的公式來處理相關係數矩陣。
2016 年
Alexander Antonov
論文 1:The Free Boundary SABR Natural Extension to Negative Rates
論文 2:FVA for General Instruments
論文 3:Backward Induction for Future Values
In all his papers there is a clear practical problem, amazing mathematics and practical implementation. I think the combination of those three elements is really quant work at its best. -- Paul Glasserman
新聞鏈接:
https://www.risk.net/awards/2442477/quant-of-the-year-alexandre-antonov
亮點:在負利率環境下的提出自由邊界 (free boundary) SABR 模型,歐式期權仍有解析解或者近似解析解,校正快,計算敏感度也沒有之前 shifted SABR 模型產生的跳躍的現象。
2015 年
Christoph Burgard
Mats Kjaer
論文:Funding Strategies, Funding Costs
The way they have approached the problem is revolutionary. They have gone back to basics and modified the Black-Scholes PDE. And because it is intuitive, it is very revealing in that you can see the cashflows in a very transparent way. -- Andrew Green
新聞鏈接:
https://www.risk.net/derivatives/2387793/quants-year-christoph-burgard-and-mats-kjaer
亮點:為業界各說各話的融資估值調整 (Funding Valuation Adjustment, FCA) 提出一個健全的理論框架。
2014 年
Michael Pykhtin
論文:Exposure under Systemic Impact
Systemic risk is at the forefront of everyone』s mind but is notoriously difficult to quantify. Pykhtin』s clear and pragmatic approach goes a long way towards setting a rigorous framework to measure and control it. -- Vladimir Piterbarg
新聞鏈接:
https://www.risk.net/awards/2320285/quant-year-michael-pykhtin
亮點:專門針對於系統重要性交易對手 (Systemically Important Counterparty, SIC) 提出系統內錯向風險 (Systemic Wrong-Way Risk, SWWR) 來量化它們違約造成的後果。
I see this as an important part of my role – communicating these technical details, This dialogue between industry and regulator is an increasingly valuable function as rules and guidelines get more technical. It』s familiar to Pykhtin – from both sides.
2013 年
Pierre Henry-Labordère
論文 1:Being Particular about Calibration
論文 2:Cutting CVA"s Complexity
It』s not complicated, actually. Using Malliavin is no harder than using the It? lemma or the Girsanov theorem. -- Pierre Henry-Labordère
新聞鏈接:
https://www.risk.net/awards/2232028/quant-of-the-year-pierre-henry-labordere-societe-generale
亮點:用法國人逆天的數學來在金融界炫耀,當然大量的減少了兩大難題的計算量,分別是「局部隨機波動率 (Local Stochastic Volatility, LSV) 模型校正」和「組合層面的信用估值調整 (Credit Valuation Adjustment, CVA) 計算」。
2012 年
Jesper Andreasen
Brian Huge
論文 1:Volatility Interpolation
論文 2
:Random Grids
There are no fundamental laws handed down from God on clay tablets. I think there is still a tendency to see the world through models, forgetting they are only as good as their implementation. -- Jesper andreasen
新聞鏈接:
https://www.risk.net/awards/2133160/quants-year-jesper-andreasen-and-brian-huge-danske-bank
亮點:1. 找到一種無套利的波動率插值方法;2. 提出一個模型校正、偏微分方程有限差分和蒙特卡洛模擬的一致離散化的想法。
2011 年
Vladimir Piterbarg
論文:Funding Beyond Discounting Collateral Agreements and Derivatives Pricing
What Piterbarg is doing is rewriting Black-Scholes post-financial crisis. After the crisis, you can』t ignore the cost of funding in any asset class or you lose money. -- Alex Langnau
新聞鏈接:
https://www.risk.net/awards/1934297/quant-year-vladimir-piterbarg-barclays-capital
亮點:對衍生品定價時引入融資成本 (cost of funding),而且這些調整可以完美的添加到整套 Black-Scholes 框架中。
2010 年
Marco Avellaneda
論文:A dynamic Model for Hard-to-Borrow Stocks
Short selling is a common scapegoat during financial crises. In 2008, the ban on short selling was also used as a form of protectionism for propping up the stock of financial firms. -- Marco Avellaneda
新聞鏈接:
https://www.risk.net/awards/1567801/quant-of-the-year-marco-avellaneda
亮點:對於難以去借 (hard-to-borrow) 來做空的股票,買賣權平價關係 (put-call parity) 不在適用。
2009 年
Lozenro Bergomi
論文:Smile Dynamics III
His idea of directly modelling the joint dynamics of the spot and variance swap volatility is theoretically sound and practically easy to implement. His quant of the year award is well deserved. -- Alexander Lipton
新聞鏈接:
https://www.risk.net/awards/1496978/quant-year-lorenzo-bergomi
亮點:提出可以控制遠期方差微笑 (smile of forward variance) 而且可以校正于波動率指數 (volatility index, VIX) 期貨和期權的模型。
2008 年
Dilip Madan
論文:Calibrating and Pricing with Embedded Local Volatility Models
The most important moment of my career was my meeting with professor Dilip Madan. He is one of the few academics that are aware that the future does not behave like the past. -- Peter Carr
新聞鏈接:
https://www.risk.net/awards/1498261/quant-year-dilip-madan
亮點:提出一個內嵌型的局部波動率模型 (embedded local volatility model) 來對波動率指數期權、股票和利率混合產品進行定價。
2007 年
Paul Glasserman
Michael Giles
論文:Smoking Adjoints Fast Monte Carlo Greeks
The adjoint method accelerates the calculation of Greeks via Monte Carlo simulation by, in essence, rearranging the order of calculations, as compared to the standard method. -- Paul Glasserman
新聞鏈接:
https://www.risk.net/awards/1498251/quants-year-paul-glasserman-and-michael-giles
亮點:用反向方法 (adjoint method) 來計算複雜衍生物的敏感度,比傳統的有限差分 (finite)、路徑微分和似然比在計算敏感度的精度不減速度卻爆升。
2006 年
Vladimir Piterbarg
論文:Time to Smile
The global skew is some sort of average of local skew. -- Vladimir Piterbarg
新聞鏈接:
https://www.risk.net/awards/1497820/quant-year-vladimir-piterbarg
亮點:提出參數平均 (parameter averaging) 的想法,將和時間有關的參數比如偏斜 (skew)、波動率轉換成和時間無關的有效 (effective) 參數,加快了複雜模型的校正速度確沒有降低校正質量。
2005 年
Philipp Sch?nbucher
論文:A Measure of Survival
Sch?nbucher is one of the most innovative researchers in credit and many of today』s practitioners have benefited from his insights. His work on CDS option pricing is typically focused and thorough, and will form the backbone of future work on the subject. -- Richard Martin
新聞鏈接:
https://www.risk.net/awards/1497632/quant-year-philipp-schonbucher
亮點:用可違約資產 (defaultable asset) 當計價物 (numeraire),在各種波動率設定下推導出類似 Black-Scholes 公式來對信用違約互換期權 (CDS option) 進行定價。
2004 年
Michael Gordy
論文:Random Tranches
Gordy』s work in portfolio credit risk is both distinguished and topical, with many of his papers being among the cornerstones of modern credit risk management practices. His work at the Federal Reserve has been highly influential with academics and practitioners alike. -- Leif Andersen
新聞鏈接:
https://www.risk.net/awards/1498479/quant-year-michael-gordy-us-fed
亮點:在符合巴塞爾大框架下,提出了一個簡單公式,能計算證券化分層所需的監管資本,解決了巴塞爾對此類產品提出過於複雜要求的監管痛點。
2003 年
Peter Carr
論文:Black-Scholes Goes Hypergeometric
Peter has contributed more fundamental ideas to the area of mathematical finance in the past couple of years than anyone I am aware of. Peter lives, eats and breathes mathematical finance. -- Keith Lewis
新聞鏈接:
https://www.risk.net/derivatives/1506232/quant-of-the-year-peter-carr
亮點:用不同的波動率設定來推廣 Black-Scholes 公式,並推導歐式期權和障礙期權的解析解。
2002 年
Richard Martin
論文:Taking to the Saddle
In credit risk modelling, he』s the most switched on person I know. -- Tom Wilde
新聞鏈接:
https://www.risk.net/awards/1506446/2002-winner-quant-year-richard-martin
亮點:用鞍點法 (saddle-point method) 代替了蒙特卡洛模擬來對損失事件建模和計算組合損失分布。
2001 年
Leif Andersen
Jesper Andreasen
論文 1:Jumping Smiles
論文 2:Static Barriers
無新聞鏈接
亮點:1. 提出跳躍擴散 (Jump-Diffusion) 模型改進局部波動率模型,因為後者生成的微笑曲線隨著時間越來越平,不符合實證觀察。2. 引進跳躍擴散模型來靜態對沖障礙期權。
2000 年
Alexander Lipton
論文:Similarities Via Self-Similarities
無新聞鏈接
到處都下載不到他的論文,甚至都很難搜索出來。聽說是關於介紹複雜衍生物的定價方法論,但是細節不清楚因此不評價。
總結
這 20 年的年度寬客和他們得獎論文主要分成市場風險 (Market Risk),信用風險及估值調整 (Credit Risk, XVA),方法論和機器學習這四大類,如下圖所示:
從上表來看,研究市場風險的趨勢在下降;研究估值調整,融資成本 (funding cost) 和保證金 (initial margin) 越來越多;研究機器學習的從今年剛開始有第一篇,按著大趨勢以後會越來越多。
亮點一評:2007 年的最佳論文 Smoking Adjoints: Fast Monte Carlo Greeks 真實好東西,這個 Adjoint 方法其實和機器學習裡面的反向傳播非常類似,這種反向求導數的方法統稱 Adjoint Automatic Differentation, AAD,在金融和機器學習中有太多應用,比如百慕大期權蒙特卡洛求敏感度,比如組合層面的 XVA,比如深度神經網路的反向傳播,只要求少量輸出對大量輸入的導數,AAD 在效率和速度上會讓你重新認識這個世界。
感嘆一聲:傳統的衍生品的定價方法不存在了,現在定個價單單看產品的風險因子完全不夠,交易對手、融資成本和保證金都會影響衍生品的價格。
世界越來越複雜,但能用簡單優雅的模型來描述他的寬客才配的起 Quant of the Year 這個稱號!
論文下載
在後台輸入
20181204
不希望某些人拿著我們的資源去收費
後台獲取方式介紹
群里已經分享了66篇乾貨
TAG:量化投資與機器學習 |