Naoki Nagai, a 36-year-old Harvard graduate who grew up in Japan, is a one-man hedge fund. For the past 16 months he has written hundreds of algorithms in much the same manner as quantitative traders in the City of London or Wall Street. But, rather than trade from a Canary Wharf skyscraper or a Manhattan boutique fund, he does so from his home in Honolulu.
In August 2006, Nagai left his job as a management consultant in Tokyo to establish a translation company, which over the next few years began to thrive. The success of his organisation, and the fact it wasn't dependent on location, gave Nagai the opportunity to reconsider his lifestyle. He chose to move from Japan to Hawaii. With its appealing climate and laid-back lifestyle, Honolulu seemed a great place to raise a family. Nagai and his wife arrived in the US in January 2014.
The Japanese trader was used to being itinerant: he had grown up in a Tokyo suburb, but studied for a Baccalaureate at an international school in South Wales before reading applied mathematics at Harvard. After college he was recruited by McKinsey as an analyst in its Tokyo office. He spent his days researching companies that operated in the complex world of semiconductors. "I wanted to be an entrepreneur," Nagai says, "and management consultancy seemed the best way to learn how to do that."
After a couple of years as a consultant, Nagai set up the translation agency, met his wife and settled down. Throughout his life he had coded as a hobby, so when he learned about a growing class of US hedge funds that traded using proprietary algorithms, he became interested. The algorithmic approach made sense to someone who saw the world in terms of data and how it might be parsed. These hedge funds were staffed by highly paid quantitative analysts, or quants, who used maths and statistics to model complex financial instruments - by leveraging the most up-to-date, detailed research and trading platforms. They also operated in a marketplace worth trillions of dollars. Nagai realised that to trade in this way, he'd have to build a tool with the same professional-grade qualities that Wall Street quantitative hedge funds such as Renaissance Technologies, PDT Partners Fund and DE Shaw used. It seemed a complex undertaking with a questionable chance of success.
Towards the end of 2014, Nagai encountered Quantopian, a Boston-based company that enables so-called retail traders - private individuals rather than institutions - to build, test and submit trading algorithms of their own invention. To submit an algorithm, it was necessary to understand the common programming language Python. Nagai set about learning and, within a month, had submitted his first algorithm. Since then, he has submitted around a dozen, coming second in the Quantopian Open on one occasion with an algorithm that had a healthy 16.87 per cent annual return.
Although he doesn't think it will happen soon, Nagai's long-term aim is to be able to live well on the profits. He's confident that his continuing study of the strategies pursued by the experts will pay dividends. "If you can study it, you can apply it," he explains.
At an elementary level, an algorithmic-trading strategy consists of three core components: entry; exit and position sizing. The margins between these components will determine the success - or otherwise - of the trade. Nagai employs the commonly used technique of arbitrage - buying and selling simultaneously in different markets to take advantage of differing prices. Any successful approach to trading is based on exploiting market inefficiencies: whether a trader is working for a top hedge fund or in their pyjamas in Silver Lake, they're fundamentally trying to do the same thing. The difference often comes down to infrastructure and data quality.
Cloud9Trader: Launched in January 2017, Cloud9Trader lets developers profit from algorithms by trading via a broker. The price data stored on its system allows a program's performance to be tested against real-world factors before live trading begins.
Quantopian: Founded in 2011 by John Fawcett and Jean Bredeche, Boston-based Quantopian provides a platform for developers to test algorithms for free. Successful applicants earn royalties when their algorithm is used. The company is managing funds for investors and is planning to make a product for institutions.
Numerai: Founded by mathematician Richard Craib in 2015, San Francisco-based Numerai attracts data scientists who develop algorithmic models using artificial intelligence. The platform helps users predict movements in the stock market and offers them the chance to win money by competing in tournaments.
Quantiacs: Founded in 2014, California-based Quantiacs gives developers a place to connect algorithms to capital from institutional investors. It supplies the data needed to test the platforms and matches hedge funds and developers. It also allows algorithm creators to share in the profits when investors use their code.
Around the time Nagai was establishing the translation business, London-based Jon Kafton was beginning an entrepreneurial journey of his own. Having worked in the City as a software developer, he had seen how banks were transitioning their trading from desktop software packages to web-based technologies. Just as other organisations, from telecos to healthcare, were moving their critical systems to the cloud, banks were migrating their core business applications to environments that were more flexible and that offered opportunities to parse data in novel and profitable ways. Having a front-row seat to the workings of the transactional economy stirred his ambition: he wanted to trade too.
Kafton investigated the types of products available, but was unable to find anything that had been designed after the beginning of the mobile era. Given the huge structural shifts that had occurred because of consumers' newfound relationships with connected devices, this surprised him.
"Amateur traders were struggling with archaic desktop software from the 90s," Kafton says. "It hadn't benefited from the current generation of digital technology that has seen enormous advances in cloud computing and user design."
Read more: The billionaire waging a war against bad science
According to Kafton, software designers and programmers were struggling to get to grips with proprietary programming languages such as Windows-based MetaTrader - "They almost seemed deliberately obscure," he says, "like 90s desktop software" - which amateur traders needed to write their algorithms. Kafton decided to develop his own proprietary programs, and he sensed an opportunity: if he couldn't find effective out-of-the-box software, surely there were others in a similar position.
In January 2017, Kafton launched his own platform, Cloud9Trader. He designed it as "an ecosystem where you engage the customer with a realistic toolset that does not try to sell them any delusion that they are going to make a lot of money fast".
Cloud9 is a new addition to the growing number of quantitative-finance platforms trying to harness the talents of amateur traders. In the way that data scientists, statisticians, physicists, computer scientists and mathematicians have taken to platforms such as Kaggle, which partners with large organisations such as General Electric to offer prizes in return for users' algorithms, so platforms such as Cloud9 and US-based Quantopian, Numerai (which raised $6 million (£7.5m) in December 2016) and Quantiacs are rewarding those who build successful algorithmic strategies on their platforms. This can take the form of prizes or licensing of the automated strategies, which are traded using institutional money from investors. Numerai says it has given $150,000 in bitcoin to contributors.
To some extent, this mirrors a hedge-fund industry trend where data-driven, quantitative funds have experienced significant growth. Investors are pulling away from traditional funds with human-led strategies, but there is more investment in quantitative funds. There was $38 billion of institutional investment into algorithmically driven hedge funds in the first quarter of 2016 alone. In October 2016, one of the biggest players, Renaissance Technologies, which has $60 billion under management, announced that it received $7 billion in investment the previous year.
The market share of platforms aimed at retail traders is tiny compared to that of the large institutions. However, it's clear that technology is offering opportunities to those beyond the maths and physics graduates from elite universities who dominate the quantitative-finance industry. Like autonomous vehicles or virus detection, machines can make real-time decisions that are faster and more accurate than those by humans.
"The time will come when no human investment manager will be able to beat the computer," David Siegel, the co-founder of quantitative fund Two Sigma, which manages $35 billion, told an investment conference in 2015.
The bet of quantitative-retail platforms is that, if extensive real-time data from financial marketplaces is available to anyone, anywhere, with the talent to write algorithms, might we be living in an era where a smart kid in Bangkok with a mathematical bent and a modest grasp of JavaScript can attract investors in the way that a Wall Street investment bank can?
One hedge-fund titan who believes in the potential of quantitative platforms for amateurs is Steve Cohen, the founder of Point72 Ventures. In 2013, he made a seed investment in Quantopian. In November 2016, Quantopian raised another $25 million in a series C round led by Andreessen Horowitz, the venture-capital firm with Facebook and Oculus VR in its portfolio.
Quantopian licenses algorithms from users, offering them a share of net profit in return. Its founder, John Fawcett, 39, was an analyst at hedge fund 033 Asset Management and founded Tamale Software, which he sold for $70 million.
In 2011, Fawcett had the idea to commoditise the process. From New York to London, hedge funds possess vast amounts of information, but the institutions are closed environments that view their data assets as something to be hoarded. Fawcett wanted to build a platform that would enable quants to conduct activities without the need for institutions. He envisioned providing mathematicians and statisticians with professional-grade tools - the infrastructure and data - they required without having to survive the rigorous process required to land a job on Wall Street or in the City. "That was not the stickiest relationship, really," he says of the quants and their employers. "I started thinking about the role of a platform within the industry and replacing this plumbing type of work, like a backtester and writing an execution engine [so that] the valuable work could be done by researchers creating new strategies."
Fawcett's play is a textbook technology gambit: remove the middle man and give individuals access to data sets and toolsets, so they are freed from institutions. His product lets individuals run historical data through an algorithm and simulate the returns. "We wanted to turn the employment model inside out in a way that benefited the talent," he says. "There are so few people who can do this well, we wanted to focus on them and create a model where they could be stars."
According to Fawcett, Quantopian has 100,000 users in 180 countries and claims Quantopian is "institutional quality". The institution is, effectively, in the browser. "If quants look at it and feel that it's as good, if not better than the tool set they have internally, that platform has really been the beacon to attract our community," he says.
Fawcett says that Quantopian has focused on removing barriers for users. It was built using Python to make it easy to use and had access to Pandas, the open-source data-analysis package designed by the statistician and computer scientist Wes McKinney, which is enormously popular with quants. Traders can build automated strategies by conducting initial research, writing algorithms, backtesting them and conducting live trading.
"We've standardised the design of an investment algorithm," Fawcett says. "We have an interface and if you write to our spec, we can test it. We can trade it with your account and we can license it from you and trade it with our account. Providing that standardisation increases people's productivity and opens up opportunities for them."
Fawcett stresses that users who have submitted more than 400,000 algorithms retain copyright over their material. "We are not trying to secretly gather all the IP and exploit people using the platform. And, selfishly, that avoids selection bias where the people who show up to write their strategies are the people who are good at it - they know they're going to create something of value, so they want to retain that. It's better for us as a company that if you create something on Quantopian it belongs to you."
Quantopian's terms of service require that submitted algorithms be tested, although the platform doesn't look at the specific code at this point. Every strategy is given a creation date. This is a crucial aspect of automated trading: establishing the data available to the author while building the strategy (in-sample) and the data that was generated afterwards (out-of-sample). The algorithms are tested using in-sample and out-of-sample data sets that examine 50 features using a top-down analysis. This means taking the big picture and breaking it into sub-systems as if the code was being reverse engineered. It's run through a simulator for a "bottoms-up" analysis, which examines the code's detail.
"We screen them and then, using the simulation, make a projection on our confidence in the algorithms' forward returns," Fawcett explains.
Read more: Brent Hoberman left lastminute.com and formed an empire. This is how he did it
Quantopian is also trying to build a product for institutional investors, as well as managing funds for investors such as Cohen, the requirements of which are very different to those of individual retail investors. Institutions look to a portfolio or fund manager for a return stream that is uncorrelated with the rest of the market. One way to do this is to remain market neutral by taking both long and short positions, meaning that the investor is insulated from the volatility of the market. Returns are generated by long positions outperforming the shorts.
"After we've evaluated the algorithm, we offer to pay the author a royalty agreement whereby they license their IP to us for use in our portfolio in exchange for a royalty on [the algorithm's] performance," Fawcett adds.
The Quantopian monetisation model is dependent on fees for managing outside investors’ money, such as those provided by Steve Cohen, using a portfolio of algorithms from those submitted, with algorithm authors receiving 10 per cent of the net profits generated from their strategy. "So much of the internet was based on consumption," Fawcett says. "I was excited about doing something that was about people being productive."
Cloud9 is built using JavaScript, which connects to brokerages via an API. Kafton maintains that anyone with even limited programing experience will find the platform "a doddle". "If you have a trading strategy, being able to turn that into code should be easy," he says. Code is static and will only work for a certain market at a given time, so programmatic traders must be willing to shift strategies. "A good strategy will target a range of market conditions, but be flexible enough to be adaptable for the changing markets," Kafton says.
This is best done by employing external parameters. If you're an investor who wants to buy when the price of a stock leaves a certain envelope - an indicator of the upper and lower price ranges of a financial instrument - you could adjust your algorithm so that, when the market becomes volatile, the envelope is wider. These external parameters occur outside of an algorithm and must be optimised using backtesting. "You have got to continually give your algorithm attention," Kafton explains. "It's not that you can forget about it, but it takes less time than sitting at home staring at charts every day."
The origins of day trading, where retail traders who buy and sell financial instruments and close out their position at the end of the day, dates back to Margaret Thatcher and the Big Bang - financial-market deregulation - in 1986. But quantitative trading is different.
Trading using an algorithm is different to live trading in one crucial way: it removes emotion. Trades occur according to pre-determined parameters established after careful consideration. They won't be swayed by market volatility and the human responses that can provoke. You could program the algorithm to enter a market at a certain point (say, when orange juice futures are at £180), exit at another (when orange juice reaches £180.50) and it will only execute those trades, no matter what else is happening in the market.
"The problem with manual trading is that we see patterns where they don't exist. It's like staring at the clouds," Kafton says. "Algo trading is more just about organising your strategy upfront so that you have got clear boundaries to your risk exposure."
Also, as with other platforms that have become destinations for those with passion and expertise, the new platforms offer something even more important: capital. For most small players, profit is squeezed from small market fluctuations. The more money you have available, the more leverage, the larger the potential returns.
One warm lunchtime in June 2016, Dan Houghton sits at a table wearing cargo shorts and eating a burrito in one of the Mexican restaurants in London that he co-owns with his business partner Eric Partaker. Houghton, a serial entrepreneur, has grown the business, which was founded in 2007. It's raised several rounds of finance, including £3.5 million on CrowdCube in December 2015. "We're the largest crowdfunded business on these kinds of platforms," Houghton says. "Other firms have raised more through crowdfunding, but they've done it direct. We're the highest on a platform."
Houghton, who has a first-class degree in mathematics and a masters in natural language processing from the University of Cambridge, has worked at several technology companies, including Skype. In his spare time, he created a text-messaging service, which he subsequently sold.
"That gave me the money to finally put a deposit on a house in London in 2015," he says. There was some money left over, which Houghton wanted to put to work. He started reading about investing and asset allocation, including The Ivy Portfolio by Mebane Faber, a quantative-analyst blogger and author who runs his own fund.
"It's about how Ivy League universities go about asset allocation," Houghton says. "Right at the end, it started to talk about a more active strategy: it's a simple quantitative strategy that takes advantage of momentum over diversification. I thought, 'Wow, you can do stuff with maths and basic algorithms, so you can grow your money faster.'"
Houghton read more books and listened to podcasts before trying out some strategies, which he built in Excel. He realised he could make money, but would need to actively oversee his investment. That doesn't work for someone who is growing a business and can only trade at night. "That's not for me," he explains. "I need to be hands-off. I run this business, I have a family."
Houghton came across a mention of Quantopian in a forum on another platform. Initially, he didn't think it was for him: building an algorithm on Quantopian required proficiency in Python, which he didn't know. "It looked a bit technical," Houghton says. "Then I realised that was the only way to go - having some kind of service where they run it for you."
He opened an account in early 2015. "The first algorithm was a vehicle for volatility futures. This strategy essentially buys volatility when it looks cheap and sells it when it looks expensive. You'll do well when markets are calm. When markets are volatile, you can see a steep drop in the value of the futures. So it's about looking for inefficiencies in the way that volatility is priced."
Before sterling began tanking against the dollar (Houghton pulled his money out when Brexit uncertainty hit the markets) he was playing with around £45,000. "You have to trade with real money," Houghton argues. "If you trade with paper, you don't really care because it's just paper. And if you don't make real trades, you don't know what the market really does when you make those trades. You have to trade something that's material enough for you to feel pain if you lose something. I made some silly errors, which obviously you don't want to do if you are trading with large amounts of cash."
Houghton now has about $20,000 that collects the Volatility Risk Premium, a strategy that rewards investors for holding volatile securities. "This has done very well in the past few months, recovering a large drawdown in the middle of the year," Houghton wrote in an email update in January 2016. "My initial capital was $15,000, less than a year ago. This is more of a fun rollercoaster of a strategy, rather than something I'd recommend to other investors. It's got an incredibly high tail risk, meaning it can be making consistently strong gains, but then out of nowhere, it can drop like a stone and lose everything. Most people don't have the stomach for it, so I'm using it to test my steel as an investor."
Read more: How an army of postmen is turning China's rural stores into the world's largest retail network
Quantitative traders work by detecting market inefficiencies. From there the process is fairly straightforward: code it up and backtest it until you've proven it. And then automate the trade and monitor it. "It's hands-off, pretty much," Houghton says. "Part of my motivation is to make money, but to also to satisfy the intellectual and competitive side of me. I really like learning about this stuff. It's really interesting. I saw Quantopian had a contest with modest prizes, but it was worth entering as they put the best ideas into their fund and give you part of the upside. I thought, 'This is brilliant, I get a share of the upside and it might make me quite rich in my spare time.'"
Amateurs such as Houghton make up a tiny fraction of the quantitative-finance industry, but the mission of Quantiacs, Cloud9, Quantopian and others is to offer a platform to anyone with the skills and willingness to participate. "You're not trying to beat hedge funds at their game, you're trying to come up with a new idea that's slightly diversified from what they're doing," Houghton says. "Obviously, the hedge fund's job is to hunt for the same thing. But what's attractive about it from an investor point of view is: if you source the ideas in a completely different way, do you end up with different ideas, which are more diversified than the pool of a few hundred people who work for big institutions? Amateurs are coming up with lots of different of ideas, many of which will be rubbish - but that's the case with most ideas."
Interested in learning more about the future of finance? Join us at WIRED Money in Studio Spaces, London on May 18, 2017. For more details and to purchase your ticket visit wiredevent.co.uk
This article was originally published by WIRED UK
