Quantum for Quants

量化交易量子化

 

The finance industry has had a long and profitable relationship with computing. It was an early adopter of everything from mainframe computers to artificial intelligence. For most of the past decade more trades have been done at high frequency by complex algorithms than by humans. Now big banks have their eyes on quantum computing, another cutting-edge technology.

 

金融业长期保持着与计算技术有利可图的关系。在采用从大型主机到人工智能的各种技术时，这个行业都走在前头。在过去十年的大部分时间里，由复杂算法驱动的高频交易的交易量已经超过了人工交易。现在，大银行又盯上了另一项尖端技术——量子计算。

 

This is the idea, developed by physicists in the 1980s, that the counter-intuitive properties of quantum mechanics might allow for the construction of computers that could perform mathematical feats that no non-quantum machine would ever be capable of. The promise is now starting to be realised. Computing giants like Google and IBM, as well as a flock of smaller competitors, are building and refining quantum hardware.

 

物理学家在上世纪80年代提出，借助量子力学种种与直觉相悖的特性，或许可以发明出量子计算机，实现非量子计算机永远无法企及的运算壮举。这个远景现在已开始逐渐成为现实。谷歌和IBM等计算巨头以及一大批较小的竞争对手都在打造和完善量子硬件。

 

Quantum computers will not beat their classical counterparts at everything. But much of the maths at which they will excel is of interest to bankers. At a conference on December 10th William Zeng, head of quantum research at Goldman Sachs told the audience that quantum computing could have a “revolutionary” impact on the bank, and on finance more broadly.

 

虽然量子计算机并不会在所有领域都击败经典计算机，但它们擅长解决的许多数学问题却让银行家甚感兴趣。在12月10日的一次会议上，高盛的量子研究负责人威廉·曾（William Zeng）对听众表示，量子计算可能给这家银行以及更广泛的金融产业带来“革命性”影响。

 

Many financial calculations boil down to optimisation problems, a known strength of quantum computers, says Marco Pistoia, the head of a research unit at JPMorgan Chase, who spent many years at IBM before that. Quantum quants hope their machines will boost profits by speeding up asset pricing, digging up better-performing portfolios and making machine-learning algorithms more accurate. A study by BBVA, a Spanish bank, concluded in July that quantum computers could boost credit-scoring, spot arbitrage opportunities and accelerate so-called “Monte Carlo” simulations, which are commonly used in finance to try to model the likely behaviour of markets.

 

摩根大通研究部门负责人、曾在IBM工作多年的马可·皮斯托亚（Marco Pistoia）说，许多金融运算归根结底都属于最优化问题，而这正是量子计算机众所周知的强项。有量子计算加持的量化投资机构希望通过加速资产定价、挖掘表现更好的投资组合、提高机器学习算法的准确性来提升利润。西班牙对外银行（BBVA）在7月完成的研究称，量子计算机可以增强信用评分、发现套利机会，并加速“蒙特卡罗”模拟——金融业广泛使用这种模拟预测市场动向。

 

Finance is not the only industry looking for a way to profit from even the small, unstable quantum computers that mark the current state of the art; sectors from aerospace to pharmaceuticals are also hunting for a “quantum advantage”. But there are reasons to think finance may be among the first to find it. Mike Biercuk of Q-CTRL, a startup that makes control software for quantum computers, points out that a new financial algorithm can be deployed faster than a new industrial process. The size of financial markets means that even a small advance would be worth a lot of money.

 

当前最先进的量子计算机仍然规模小且不稳定，但即便如此，也不是只有金融业在试图利用它获利——从航空航天到制药等众多行业都在追寻“量子优势”。但有理由认为金融业可能是捷足先登的行业之一。Q-CTRL是一家为量子计算机开发控制软件的创业公司，公司创始人迈克·比埃库克（Mike Biercuk）指出，部署一种新的金融算法可能比部署新的工业流程要快。金融市场的规模如此庞大，就算小小的进步也会价值连城。

 

Banks are also buying in expertise. Firms including BBVA, Citigroup, JPMorgan and Standard Chartered have set up research teams and signed deals with computing firms. The Boston Consulting Group reckons that, as of June, banks and insurers in America and Europe had hired more than 115 experts – a big number for what remains, even in academia, a small specialism. “We have more physics and maths PhDs than some big universities,” jokes Alexei Kondratyev, head of data analytics at Standard Chartered.

 

银行也在为专业知识掏腰包。包括西班牙对外银行、花旗集团、摩根大通和渣打银行在内的许多公司已经成立了研究团队，并与计算公司签署协议。波士顿咨询集团估计，截至6月，欧美的银行和保险公司已经聘请了超过115名专家。鉴于量子研究即使在学术界也仍属小众，这个数字可谓相当惊人了。渣打银行的数据分析负责人阿列克谢·康德拉特耶夫（Alexei Kondratyev）打趣说：“我们这里的物理和数学博士比一些大型大学还多。”

 

Startups are exploring possibilities too. Enrique Lizaso of Multiverse Computing reckons his firm’s quantum-enhanced algorithms can spot fraud more effectively, and around a hundred times faster, than existing ones. The firm has also experimented with portfolio optimisation, in which analysts seek well-performing investment strategies. Multiverse re-ran decisions made by real traders at a bank. The job was to choose, over the course of a year, the most profitable mix from a group of 50 assets, subject to restrictions, such as how often trades could be made.

 

创业公司也在探索其中的种种可能。平行宇宙计算公司（Multiverse Computing）的恩里克·利萨索（Enrique Lizaso）认为，他的公司经量子增强的算法可以更有效地甄别欺诈，大概比现有算法快约100倍。该公司也测试投资组合优化，帮助分析师找到表现优异的投资策略。它重演一家银行的真人交易员所做的决策。这项工作要求在一年时间内，在交易频率等限制条件下，从50项资产中选择盈利能力最强的组合。

 

The result was a problem with around 10^1300 possible solutions, a number that far outstrips the number of atoms in the visible universe. In reality, the bank’s traders, assisted by models running on classical computers, managed an annual return of 19%. Depending on the amount of volatility investors were prepared to put up with, Multiverse’s algorithm generated returns of 20-80% – though it stops short of claiming a definitive quantum advantage.

 

这样就产生了一个约有10^1300个可能解的问题，这个数字远远超过可见宇宙中的原子总数。在现实中，该银行交易员在经典计算机上运行的模型的辅助下实现了19%的年回报率。而根据投资者愿意承受的不同波动幅度，平行宇宙公司的算法产生了20%到80%不等的回报——尽管还没有达成绝对的量子优势。

 

Not all potential uses are so glamorous. Monte Carlo simulations are often used in regulatory stress tests. Christopher Savoie of Zapata, a quantum-computing firm based in Boston, recalls one bank executive telling him: “Don’t bring me trading algorithms, bring me a solution to CCAR (an American stress-test regulation). That stuff eats up half my computing budget.”

 

并非所有的潜在应用都如此引人入胜。蒙特卡罗模拟常用于监管压力测试。总部位于波士顿的量子计算公司Zapata的克里斯多夫·萨瓦（Christopher Savoie）记得一位银行高管曾对他说：“不用给我交易算法，能想办法让我通过CCAR（美国一项压力测试法规）就行。这玩意吃掉了我一半的计算预算。”

 

All this is promising. But quantum financiers acknowledge that, for now, hardware is a limitation. “We’re not yet able to perform these calculations at a scale where a quantum machine offers a real-world advantage over a classical one,” says Mr Biercuk. One rough way to measure a quantum computer’s capability is its number of “qubits”, the analogue of classical computing’s 1-or-0 bits. For many problems a quantum computer with thousands of stable qubits is provably far faster than any non-quantum machine that could ever be built – it just does not exist yet.

 

所有这些前景无限。但采用量子技术的金融家也承认，目前硬件仍是软肋。“现在量子计算机运算的规模还是太小，不足以在真实世界里实现对经典计算机的优势。”比埃库克说。对量子计算机性能的一种粗略的衡量方法是看它的“量子比特”数量，量子比特类似于经典计算机中的1或0。对于许多计算问题来说，一台拥有数千个稳定量子比特的量子计算机理论上要比任何能造出来的非量子机器都快得多——只是它尚未出现。

 

For now, the field must make do with small, unstable devices, which can perform calculations for only tiny fractions of a second before their delicate quantum states break down. John Preskill of the California Institute of Technology has dubbed these “NISQs” – “Noisy, Intermediate-Scale Quantum computers”.

 

目前，量子计算只能将就着使用不稳定的小型设备，它们的运算维持时间远远不到一秒，然后其精细的量子态就会坍塌。加州理工学院的约翰·裴士基（John Preskill）给这些设备取名“NISQ”，即“带噪声的中等规模量子计算机”。

 

Bankers are working on ways to conduct computations on such machines. Mr Zeng of Goldman pointed out that the computational resources needed to run quantum algorithms have fallen as programmers have tweaked their methods. Mr Pistoia points to papers his team has written exploring ways to scale useful financial calculations into even small machines.

 

银行家们正想方设法在这样的机器上完成运算。高盛的威廉·曾指出，随着程序员调整了方法，运行量子算法所需的计算资源已经缩减。皮斯托亚称他的团队撰写了一些论文，探讨如何缩小实用金融计算的规模，使之可以在小型机器上运行。

 

And at some point those programmers will meet hardware-makers coming the other way. In 2019 Google was the first to demonstrate “quantum supremacy”, using a 53-qubit NISQ machine to perform in minutes a calculation that would have taken the world’s fastest supercomputer more than 10,000 years. IBM, which has invested heavily in quantum computing, reckons it can build a 1,000-qubit machine by 2023. Both it and Google have talked of a million qubits by the end of the decade.

 

程序员和硬件制造商相向而行，总有一天双方可以汇合。2019年，谷歌率先展示了“量子霸权”，用53个量子比特的NISQ机器在几分钟内完成了世界上最快的超级计算机需要一万多年才能完成的计算。斥巨资研究量子计算的IBM估计自己到2023年可制造出有1000个量子比特的计算机。两家公司都谈到在十年内可以达到一百万量子比特。

 

When might the financial revolution come? Mr Savoie thinks simple algorithms could be in use within 18 months, with credit-scoring a plausible early application. Mr Kondratyev says three to five years is more realistic. But the crucial point, says one observer, is that no one wants to be late to the party. One common worry is that whoever makes a breakthrough first may choose to reap the rewards in obscurity, rather than broadcast the fact to the world. After all, says Mr Biercuk, “that is how high-frequency trading got started”.

 

这一金融革命究竟何时到来？萨瓦认为18个月内就可以部署简单的算法，早期的应用很可能是信用评分。康德拉特耶夫则认为，在三到五年内实现可能更加现实。但一位观察人士说，关键在于没人希望落在后面。大家普遍担忧的是，无论谁首先取得突破，都可能不会公之于众而是选择闷声发大财。毕竟，比埃库克说，“高频交易就是这么来的”。

英文、中文版本下载：http://www.yingyushijie.com/shop/source/detail/id/2587.html