The Future of Portfolio Returns and Wall Street Traders

This post explores the benefits of machine learning platforms on the trading business of Goldman Sachs and the long-term effects of these platforms on human capital management and the ability of firms to retain their competitive advantage in the financial services industry.

Technology has and continues to provide financial services firms, including Goldman Sachs, a significant edge in the marketplace. Some experts believe that the capital markets industry stands to gain the most from the applications of AI: both supervised and unsupervised machine learning.1 Firms generate massive amounts of data and “how that data gets harnessed, analyzed, and used across the value chain will increasingly not be up to humans but rather machines.”2

Machine learning will be an invaluable tool in tackling process improvement across Goldman’s trading business. Currently, traders utilize various trading applications to manually execute a number of critical tasks, including: processing trades, assessing and calculating risk metrics, and modeling optimal portfolios. The immense capability of machine learning to analyze millions of data points simultaneously and at high speeds and apply these learnings in real time will have significant implications for Goldman and its clients. Some of these benefits include higher returns, cost savings, and increased capacity for trading professionals. 1,2 The various applications of machine learning across trading are detailed below.
Machine learning will predict the level of systematic and idiosyncratic risk at a much faster pace than humans. Additionally, these tools will be able to quickly react to and optimize portfolios for various market events.3,4 According to experts, “deeper” machine learning applications will be able to utilize historical data sets to predict, with the highest degree of accuracy, the range of outcomes for any given market security. Machine learning will also be able to aggregate data from informal sources such as social media and blogs to gain “a previously unattainable level of insight into a stock’s trading ability.” These insights will then be utilized to model out highly sophisticated “what-if” scenarios to construct the most optimal portfolio for any given set of constraints.4

Machine learning will also harness the power of speech recognition. Experts conclude that, “in the future, speech recognition may be able to tell the intent, sentiment or even urgency of speech, not just the words themselves.”4 This “intent” will be utilized to execute trades “in a fraction of the time it takes to do an electronic transaction.”4

Over the past several years, Goldman has been intensely building and internal AI Team to improve performance and cut costs.2 But in the absence of these sophisticated machine learning tools, trading professionals will continue to utilize an in-house application called Securities DataBase or “SecDB” in the short- and medium-term. Known as Goldman’s “secret sauce,” SecDB is used to (i) track securities and their historical performance under various scenarios; (ii) model performance of these securities under future scenarios; and (iii) determine aggregate risk these securities introduce to various portfolios.5

While SecDB is a powerful tool in computing and analyzing data, the speed at which inputs to the system are provided and the speed at which outputs of the system are analyzed and applied are constrained by human capacity and are subject to a high-degree of human error.5,6 In a fast-paced capital markets environment where pricing inefficiencies exist for very short periods of time, speed is critical in gaining opportunities to generate alpha or outsized returns. Though Goldman’s quant team is continuously focused on improving SecDB, introducing new variables or new sources of information is a highly iterative process and undergoes significant testing to ensure outputs are accurate and useable.

Technology remains a key competitive advantage for financial services firms and a “rat race” has commenced as notable competitors are also focused on rapidly recruiting experts to develop a viable proof of concept. Speed to market will be key in determining “winners and losers,” as banks with well built and implemented machine learning applications will be in a position to win over a large number of clients, while laggards may struggle to survive. AI has seen a burgeoning in FinTech with start-up companies developing a gamut of solutions across financial services. Goldman would be well served in outright acquiring IP or structuring exclusive licensing agreements with start-ups to potentially box out competitors.

Goldman must evaluate the consequences of utilizing legacy systems to build its machine learning platform. It must ensure that inherent biases across SecDB are identified, as machine learning “algorithms are not natively intelligent,”7 and learn from data that is being inputted. Additionally, the Firm must also think through the extent to which a trader’s “gut feeling” or “intuition” relative to the movement of certain securities impacts performance and how an environment of evolving regulation must be accounted for in the development of these tools.

As we continue to follow this evolution of machine learning, what role will humans play in a scenario of complete trading automation? If proven machine learning solutions become available to the masses, what role will banks play? Will the banking industry experience consolidation in the long-term as firms lose their competitive advantages?

[Word Count: 798]

1Butcher Dan, “Goldman Sachs has created an elite tech team to tackle AI, big projects,” efinancialcareers, November 16, 2017, https://news.efinancialcareers.com/us-en/301350/goldman-building-new-rd-engineering-group-hiring-ai-team, accessed November 2018.

2Dickinson Claire, “AI the ‘big winner’ as banks and fund managers dig deep on tech,” Financial News London, August 16, 2017, https://www.fnlondon.com/articles/ai-the-big-winner-as-banks-and-fund-managers-dig-deep-on-tech-20170816?mod=article_inline, accessed November 2018.

3Bharadwaj Raghav, “Artificial Intelligence at Investment Banks – 5 Current Applications, techemergence, November 7, 2018, https://www.techemergence.com/artificial-intelligence-at-investment-banks-5-current-applications/, accessed November 2018.

4Coles Terri, “How Trading Systems will Shake up Wall Street,” IT Pro Today, January 12, 2018, https://www.itprotoday.com/machine-learning/how-ai-trading-systems-will-shake-wall-street, accessed November 2018.

5Baer Justin, “Understanding SecDB: Goldman Sach’s Most Valued Trading Weapon,” The Wall Street Journal, September 7, 2016, https://www.wsj.com/articles/understanding-secdb-goldman-sachss-most-valued-trading-weapon-1473242401, accessed November 2018.

6Baer Justin, “Goldman Sachs Has Started Giving Away Its Most Valuable Software,” The Wall Street Journal, September 7, 2016, https://www.wsj.com/articles/goldman-sachs-has-started-giving-away-its-most-valuable-software-1473242401, accessed November 2018.

7Press Gill, “These Banks are Using AI to Help Their Customers Manage Their Finances,” Forbes, September 12, 2018, https://www.forbes.com/sites/gilpress/2018/09/12/these-banks-are-using-ai-to-help-their-customers-manage-their-finances/, accessed November 2018.

Previous:

NYC BigApps: Crowdsourcing Civic Innovation

Next:

Baidu: Using Machine Learning for Voice Cloning to Get Closer to Consumers…All In Just 3.7 Seconds!

6 thoughts on “The Future of Portfolio Returns and Wall Street Traders

  1. This was an interesting read that questioned the longevity of arguably one of Wall Street’s most powerful and enviable financial institutions, and raised the question of their survivability prospects in the age of Artificial Intelligence. I have always believed that ML/AI technologies are prime disruptors in the financial services world (especially on the trading, and less so on the banking sides of these businesses) given the repetitive nature of their jobs, and the gains that can be made in making better decisions. The author highlights several benefits of AI/ML and how Goldman is positioning itself to improve its “secret sauce” SecDB in a way that they are effectively disrupting themselves before anyone else does so. At the same time, use of this old system as a framework for the development of a new and improved “smarter” system highlights risks that the innovation would not be as radical as Goldman expects. We saw an opposite approach to the build-out of AI at IBM Watson where they pushed to start from a clean-slate, resulting in better results – which raises the question: Should Goldman run a skunkworks designed lab to leap-frog the current approach of small iterative improvements in its improvements to its algorithm?

  2. The advance of the robots is a big worry of top traders at bulge bracket banks. While they may increase effenciency and increase returns at clients, a lot of bankers are worried that it will be a race to the bottom, and actually increase risk in the system.

    The perfect trade exists when there are direct and executable price mismatches in the market. For example, take two exactly identical pairs of shoes were for sale directly accross the street from one another, but one is priced at $10, and one is priced at $20. If you were able to buy the $10 shoes, walk accross the street and sell them to customers in the other store for $20, you have a riskless, perfect trade, or arbitrage opportunity. Electronic trader finds these market anomolies and exploits them, even if they last for only a fraction of a second. Traders argue that these traders are the safest way to get returns, or maximize alpha. If everyone employs machine trading, these opportunities wont exist anymore, and traders will start looking to trade more on their forecasts or guesses of the future, which is inheritly more risky.

  3. Great read. Seeing the growing trend of Wall Street pushing for and someday relying upon AI will be interesting to watch. Especially in terms of how it impacts efficiency, expected returns, and the role of traders within the system. When I read this, I thought of Watson identifying Toronto as a US City. A single misstep by AI of this nature, while understandable given the logic within the algorithm, could have major financial ramifications when used in trading. I wonder how firms will be able to address these risks and satisfy potential customer concern about the use of AI, beyond just in forecasting.

  4. This is a hot topic in the industry. Goldman and J.P. Morgan are the two companies that are heavily investing behind machine learning and open platform technologies. Most of the other banks say they are investing, but they are either too far behind or it’s not necessarily the #1 priority for them. I think the true competitive advantages of Goldman’s trading department is 1) its scale and 2) its ability to manage risk. Goldman is now opening up SecDB to its clients, and believes it can win more business from clients if it opens up its tech platform.

    What I disagree with is that I don’t think many of these banks will lose their competitive advantages over the medium-term. Most of the banks make the majority of their money from other business lines including traditional investment banking, asset management, trust, and S&L. Trading is highly asset intensive, given the fact that you have to hold inventory. With Fed and Basel minimum capital requirements, the ROAs of this business had shrunk, and most banks have re-focused their efforts. The true competitive advantage of a bank is its scale, its ability to manage risk, and its brand/relationship with clients and customers.

  5. Interesting read!

    Your questions about what will happen if standard ML applications become available to the masses for investing purposes is a crucial one. One of the issues that I grapple with ML in investing is to what extent will the penetration in ML/automated trading/passive investing erode alpha vs. enabling investors to capitalize market dislocations to drive alpha? Consider the following extreme scenario — if the overwhelming majority of the market is using the same ML platforms, then they will be guided to the same investment opportunities, thus resulting in more efficient price discovery and lack of differentiation in investment theses that reduces instances of alpha. In this scenario, having a differentiated (i.e., not machine/algorithmic) approach will be the only way to find alpha, as it will have been efficiently bid-away for all opportunities that ML has identified.

    Howard Marks — the legendary investor — wrote a recent memo about the role of humans in investing going forward. I highly suggest reading it to help flesh out your open items, enjoy!

    https://www.oaktreecapital.com/docs/default-source/memos/investing-without-people.pdf

  6. Great essay! I think we’ll start to see a few different classes of trading oriented AI emerge over the next 5 years. You talk about current AI’s latency challenges – I don’t see AI ever being quicker than current intraday algos. I think AI will bifurcate into tools that are centered around developing intelligence around general equity analysis (of the buy,hold,sell variety) and intraday specific tools. For the intraday tools, rather than something that is actively managing trading – I see AI as a tool that can augment the capabilities of a trader. At IBM we were working on a voice to code product, meaning a piece of AI that would translate natural speech (you talking to it) into code. I thought a great application of this would be enabling non-tech savvy traders to write trading algos on the fly.

Leave a comment