{"id":33173,"date":"2018-11-13T16:19:19","date_gmt":"2018-11-13T21:19:19","guid":{"rendered":"https:\/\/digital.hbs.edu\/platform-rctom\/submission\/billionaire-robots-machine-learning-at-renaissance-technologies\/"},"modified":"2018-11-13T16:19:19","modified_gmt":"2018-11-13T21:19:19","slug":"billionaire-robots-machine-learning-at-renaissance-technologies","status":"publish","type":"hck-submission","link":"https:\/\/d3.harvard.edu\/platform-rctom\/submission\/billionaire-robots-machine-learning-at-renaissance-technologies\/","title":{"rendered":"Billionaire Robots: Machine Learning at Renaissance Technologies"},"content":{"rendered":"
\"\"<\/a>
The “Flash Crash”: Dow Jones Industrial Average on May 6, 2010<\/em><\/figcaption><\/figure>\n

Michael Lewis chronicles the need for speed in his book \u201cFlash Boys\u201d, tracing the rise of high-frequency trading (\u201cHFT\u201d) as one of the most profitable public equity investment methodologies of the 21st<\/sup> century. Increasingly, however, trading speed is no longer sufficient to yield differentiated return \u2013 transaction time that was once measured in seconds is now monitored in nanoseconds. As more companies have built the technological infrastructure to compete in the HFT space and competitors have begun to scale, industry leaders have turned to predictive programming and, specifically, machine learning to enhance their ability to forecast stock movements to buy and sell accordingly. Renaissance Technologies (\u201cRenTec\u201d) is one of the most successful hedge funds to dominate the inception of quantitative trading, producing unrivaled returns of 71.8% annually from 1994 through mid-2014.1<\/sup><\/p>\n

Success in high-frequency trading was once solely based on geographic proximity to the NYSE and Nasdaq data centers in northern New Jersey; in fact, hedge funds would pay maintenance staff to place their servers slightly closer than their competitors\u2019 (the staff later dubbed this \u201cproximity services\u201d or \u201ccolocation\u201d).2<\/sup> The arms race for speed intensified throughout the early 2010s, culminating in the construction of Spread Networks\u2019 fiber-optics tunnel, which bridged Chicago and New York using low latency cable.3<\/sup> By the mid-2010s, HFTs comprised approximately 50-60% of public equity trading by dollar volume.4 <\/sup>Trading speed became necessary but not sufficient to produce differentiated returns. Funds turned to algorithmic trading, which uses machine learning to \u201ctrain\u201d computer models on past price data in order to predict future price movements. Machine learning and artificial intelligence are particularly suited to trading due to the volume of information available for analysis; pricing models can be trained on extensive stock market data with signals ranging from news articles to historical price movements.5,6<\/sup> Machine learning is being applied to liquid trading in two ways: first, using \u201creinforcement learning and neural networks\u201d to increase the predictive power of the computer models, and second, to determine optimal portfolio allocation to different types of securities.7,8 <\/sup><\/p>\n

RenTec has long been at the forefront of the deployment of machine and deep learning tools in trading. Its founder, James Simon, has focused on differentiating the firm by hiring investment professionals with doctoral degrees, particularly those with a science or mathematics background.9 <\/sup>Co-founder Howard Morgan explains, \u201c[RenTec doesn\u2019t] want people to have experience with trading; they want people who have math, stats, physics, biochemistry, or computer science backgrounds.\u201d10<\/sup> The firm has also focused on creating a work environment conducive to the open exchange of ideas, much more like a research lab than a traditional hedge fund. Simon himself explains, \u201cno compartmentalization \u2026 you don\u2019t get paid just on your work.\u201d11 <\/sup>Recognizing that employees are the key asset for his firm, Simons has reduced employee turnover to almost zero using equity compensation incentives and lifestyle benefits.12 <\/sup>Unlike most other hedge fund managers, who grow assets indiscriminately due to fee structure incentives (many earn 2% of AUM regardless of fund performance), Simons also realized that there are diminishing returns to taking on capital due to constraints on deployment opportunities; the fund is capped at ~$10bn and is mostly restricted to employees.13<\/sup><\/p>\n

Although RenTec has been an undisputed leader in the emergence of quantitative trading, other funds are not far behind. As traditional barriers to entry have diminished due to cheaper trading infrastructure, new entrants have flooded the industry in search of outsized profits. Many have begun to mimic RenTec\u2019s strategy of hiring non-finance PhDs. In order to preserve its competitive advantage, RenTec has begun and should continue to look at other asset classes and strategies \u2013 for example, they were one of the first to begin trading German futures.14<\/sup> With each new asset class comes a new set of risks, which have often produced losses at other firms that have attempted to diversify. Unlike other strategies, quantitative trading does not require human intervention. Since the computers trade large volumes quickly, this can produce unexpected results. The 2010 \u201cflash crash\u201d sent some stocks plummeting or skyrocketing inexplicably only to bounce back moments later.15<\/sup> RenTec should focus on honing monitoring systems that would prevent illogical and extreme trading results. In the wake of this catastrophe, regulators shown a spotlight on the previously unknown world of quantitative trading. RenTec should work to understand potential regulation that could significantly reduce the profitability of their strategies.16<\/sup><\/p>\n