Two Sigma Investments: Will Machines Take Over the Stock Market?

With over 1,500 employees, including 200+ PhDs, Two Sigma runs systematic hedge fund strategies that rely on machine learning to extract insights and predictive signals in a fraction of the time the human brain can to deliver superior alpha with less risk than traditional hedge funds. Is this competitive advantage sustainable?

Since founding Two Sigma Investments in 2001, statistician and computer scientist co-founders John Overdeck and David Seigel have sought to reinvent the investment management industry through the use of machine learning, distributed computing, and other technologies to develop cutting-edge systematic hedge fund strategies that deliver superior risk-adjusted returns for its clients. [1] Having been a client of Two Sigma’s, I wonder how accurate Two Sigma’s predictive algorithms can be in an uncertain market backdrop and as unprecedented events occur that machines have never encountered.

The IDC estimates that the amount of digital data the world produces will reach 44 zettabytes (trillions of gigabytes) by 2020, an amount so big that if it was all put in iPad Air tablets, the stack would reach from earth to the moon more than six times over. [2] This explosion of alternative data has created an arms race among hedge funds hiring data scientists to dive into pools of data and use machine learning algorithms to extract insights or predictive signals in a fraction of the time and with greater accuracy than the human brain. [3] In an industry facing heightened competition, dulled growth, volatility, and significant fee pressure, $50 billion Two Sigma differentiates itself among hedge funds by utilizing machine learning and the scientific method to quickly forecast market moves to enhance returns, control risk, and arbitrage information about stocks. [4] In an industry ripe for disruption, Two Sigma has a significant first-mover and competitive advantage with its sophisticated trading models that use world-class computing power to identify trading patterns that exploit inefficiencies and identify alpha opportunities in the public markets.

To stay ahead of the competition and address the need for superior returns and adaptation to evolving market conditions, Two Sigma has taken a number steps to remain at the forefront of innovation over the short and medium-term. Over 70% of Two Sigma’s 1,400 employees come from outside the finance industry and the majority are focused on research and development to perfect the platform through coding, modeling, and engineering. [5] Two Sigma heavily invests in machine learning systems to analyze more than 10,000 data sources using 75,000 CPUs and 35 petabytes of data. [6] They also invest in natural language processing capabilities to analyze news and video content. [7] Two Sigma recently made a minority investment in Crux Informatics, a startup that specializes in cleaning up, transforming, and reformatting giant data sets, which must be done before Two Sigma can apply data science to extract signals and value out of it. [8] The firm has focused on hiring top AI, machine learning, and quantitative PhD talent to build and improve models. For instance, they hired the former VP of Research and Special Initiatives at Google as the chief technology officer to oversee the engineers and hired the senior staff research scientist from Google on the “Brain” team doing neural networks and machine learning R&D, among other top talent in the tech community. [9] Two Sigma has traded over 300 million shares daily over the last 14 years, and through this constant feedback loop, has improved and iterated on its machine learning algorithms to be more accurate and predictive in the ever-changing markets. [10]

In the short and medium term, I would suggest that Two Sigma continue to acquire start-ups with greater computing power to accurately dissect data and language to detect signals while cutting through the noise in alternative data. There are limitations to the if-then Boolean logic upon which most machine learning is based, which is why Two Sigma needs to continue to hire top mathematicians and investors to connect non-linear information that the machine is not yet capable of analyzing, while investing in R&D to develop these capabilities for their machines. [11] I suggest they do this through open innovation and more tournaments like Halite. [12] In addition, Two Sigma should invest in R&D to help machines react to unprecedented market movements and volatility, which doesn’t have historical data to extract patterns from. In addition, a lot of investors are wary of “black box” investing where machines make investment decisions, so Two Sigma should publish their audited results to build trust from the general public and potential new clients. In the medium-term, I would suggest they offer their machine learning algorithms to other fields such as healthcare, education, and logistics to serve the broader community.

Two open issues that merit further conversation are:

1.      Will machines ever fully displace humans from the investment process or will a human’s judgement always be necessary?

2.      If all hedge funds migrate to employ machine learning algorithms to trade, will there be any inefficiencies left and any alpha to extract in the market?

(794 words)

[1] Two Sigma Investments, “About”, https://www.twosigma.com/about/, accessed November 2018.

[2] Joel Weber, Bloomberg Finance LP, “Financial Market Regulatory Wire”, September 13, 2017, https://search-proquest-com.ezp-prod1.hul.harvard.edu/businesspremium/docview/1940999256/9D261C33EEC64326PQ/4?accountid=11311, accessed November 2018.

[3] https://global-factiva-com.prd2.ezproxy-prod.hbs.edu/ga/default.aspx, accessed November 2018.

[4] Christine Williamson, Pensions & Investments, Chicago, Vol. 46, Iss.8, “Race is on to Grab Most Possible From Machine Learning: Managers Speed Ahead to Find New Ways to Make Technology Work for Them”, April 16, 2018, https://search-proquest-com.ezp-prod1.hul.harvard.edu/businesspremium/docview/2027208748/9D261C33EEC64326PQ/1?accountid=11311, accessed November 2018.

[5] Gillian Kemmerer, Absolute Return, London, February 10, 2016, “Firm of the Year Nominee: Two Sigma Investments”, https://search-proquest-com.ezp-prod1.hul.harvard.edu/businesspremium/docview/1773252046/9D261C33EEC64326PQ/7?accountid=11311, accessed November 2018.

[6] Nathan Vardi, Forbes, October 19, 2015, https://www.forbes.com/sites/nathanvardi/2015/09/29/rich-formula-math-and-computer-wizards-now-billionaires-thanks-to-quant-trading-secrets/#1639d7f46712, accessed November 2018.

[7] Gillian Kemmerer, Absolute Return, London, February 10, 2016, “Firm of the Year Nominee: Two Sigma Investments”, https://search-proquest-com.ezp-prod1.hul.harvard.edu/businesspremium/docview/1773252046/9D261C33EEC64326PQ/7?accountid=11311, accessed November 2018.

[8] https://global-factiva-com.prd2.ezproxy-prod.hbs.edu/ga/default.aspx, accessed November 2018.

[9] Gillian Kemmerer, Absolute Return, London, February 10, 2016, “Firm of the Year Nominee: Two Sigma Investments”, https://search-proquest-com.ezp-prod1.hul.harvard.edu/businesspremium/docview/1773252046/9D261C33EEC64326PQ/7?accountid=11311, accessed November 2018.

[10] Alexander Osipovich, Wall Street Journal, May 10, 2017, “The Secretive Firm Set to Expand its Retail Options: Two Sigma Securities”, https://www.wsj.com/articles/the-secretive-firm-set-to-expand-in-retail-options-two-sigma-securities-1494446194, accessed November 2018.

[11] Mark Melin, ValueWalk, February 20, 2017, “Are Machine Learning Hedge Funds Selling a Fantasy?”, https://search-proquest-com.ezp-prod1.hul.harvard.edu/businesspremium/docview/1870005327/26367CEB4BE04596PQ/3?accountid=11311, accessed November 2018.

[12] https://halite.io/, accessed November 2018.

Previous:

The Election Security Challenge: Why Voting-Machine Vendors Resist Open Innovation

Next:

Customization vs. Scale – How BMW Imagines the Future of Cars

Student comments on Two Sigma Investments: Will Machines Take Over the Stock Market?

  1. I find your second question to be extremely thought provoking. If the stock market were to entirely be transitioned to machines and machine learning/AI, it would in theory leave very little room for “seeking alpha”. One possibility that could lend to differences in machine learning results, is the bias that is encoded in the machine learning process. Each algorithm could be coded differently and would render differing results. The problem with this is that once one person discovers the code that was entered, it can easily be replicated.

  2. Interesting article, thanks for sharing!

    While Two Sigma seems to have developed models which have historically outperformed, there are hundreds of other systematic hedge funds (with PhD employees) searching for the same signals which Two Sigma is trading on. For that reason, I do not believe the competitive advantage is sustainable.

    To your question on if machines will every fully displace humans, I believe it’s highly dependent on the type of job:
    • Barriers in the private markets (i.e., lack of data for machines to train on) will keep humans employed so long as there is information asymmetry.
    • In the public markets, I expect the transition to (mostly) machines to happen much quicker.
    • Jobs that require negotiation (i.e., structuring of esoteric securities) will be more defensible.
    • Back/middle office jobs might be most at risk, due to automation and distributed ledger technology.

  3. Your second question is very interesting and leads me to believe that the market should be entirely efficient if all hedge funds are using machine learning algorithms to trade. In this case, hedge funds will need to use their machine learning technology and algorithms as their competitive advantage over other funds in order to extract any sort of value. This is an interesting concept to think about because trading will ultimately become very similar to the tech industry since firms with the best technology will have an advantage in the market. In this case, the skillsets of the traders will need to be entirely different than they are today and traders with strong technical and coding skills will be highly sought after. I am curious to see how other hedge funds begin to use machine learning and who they will hire in order to strengthen their technology platforms.

  4. I’ll take a stab at the second question you pose (If all hedge funds migrate to employ machine learning algorithms to trade, will there be any inefficiencies left and any alpha to extract in the market?). Not every hedge fund will have the same algorithm, therefore, funds that can put the best algorithms into their systems will, in theory, perform better. Also, humans still need to interpret the data the machines churn out, so a firm can gain an advantage in that aspect.

    Very interesting post!

Leave a comment