Two developments in technology, blockchain and deep learning, have implications for securities trading regulation. The two technologies are different in scope and purpose and will raise different issues for securities regulators. Both demonstrate how technological advances in the trading area can outpace current rules and regulations and cause regulators to rethink how to handle so-called “disruptive” technologies without impeding new structures and ideas.
Much has been written over the past year about blockchain (i.e., distributed ledger technology) and the challenges that it poses for financial regulators in determining how this technology fits into the current regulatory structure. The vast majority of analysis of this issue has focused on initial coin offerings and digital currencies, particularly the legal status of these instruments. As noted in a well-written FINRA report from January 2017 on blockchain, however, this technology poses issues for regulators in a variety of areas. The potential impact of blockchain on trading regulation is one area that raises many interesting issues.
Blockchain, a form of distributed ledger technology (“DLT”), at some point will enable market participants to post quotes and consummate trades without intermediation through traditional participants such as exchanges, automated trading systems (“ATSs”) or even registered broker-dealers. While DLT has more potential use for a fragmented market such as fixed income or illiquid securities, it also could be used in OTC trading of equity securities. This will raise legal status questions for the participants and providers of the DLT as it is highly likely that regulators will want some regulated entity to be involved in the DLT trading process.
Legal status questions will include, for example, whether regulators view a DLT-enabled trading mechanism to be an ATS or an exchange. In addition, will regulators require a broker-dealer to operate a DLT quote and trade mechanism? Aside from legal status issues, DLT also will implicate transparency regulations as quotation and trade activity over a distributed ledger network might be at odds with current transparency regulations. There are many other regulations that will not accommodate neatly trading via DLT, such as Regulation NMS. For example, Regulation NMS requires trading centers to prevent trade-throughs of protected quotations of NMS stocks. Will a trade in an NMS stock conducted through private negotiations via DLT be subject to Regulation NMS’ trade through rule? Moreover, settlement of transactions and transfers of security ownership via blockchain will pose its own set of issues for securities regulators.
Deep learning is another technological advance that has important implications for securities regulation. While not as advanced in its implementation as blockchain, it is only a matter of time before it becomes a tool for certain sophisticated traders, such as hedge funds and proprietary trading firms. In simple terms, deep learning is a mixture of machine learning and artificial intelligence (“AI”) that involves learning data representations and patterns using simulated neural networks. Deep learning necessitates access to huge amounts of data and immense computing power.
Once deep learning becomes a more widely used technology among sophisticated traders, regulators will have to determine how to examine and oversee trading based upon deep learning. FINRA has taken some steps in this direction. In testimony by FINRA President and CEO Robert Cook on September 7, 2017, before the House Subcommittee on Capital Markets and Securities, Mr. Cook discussed technology-driven innovation in regulation. He mentioned the use of cloud technology to improve FINRA’s ability to surveil the markets and conduct examinations. He also mentioned that FINRA has begun to explore the use of machine learning to look for suspicious trading patterns.
The dilemma that FINRA and the SEC face in surveilling and examining trading based on deep learning will be similar to the challenges regulators faced in the emergence of high frequency trading. Specifically, regulators will have to ascertain how to differentiate legitimate trading strategies based upon technological advances from abusive conduct such as spoofing and layering. At the same time, regulators may find it difficult to discern whether a deep learning strategy intended to engage in abusive conduct or the algorithm unintentionally “learned” such behavior thorough analysis and reaction to the data sets it reviewed. The use of deep learning for trading strategies almost certainly will develop before the regulators obtain the capabilities to accomplish this task. It will be important for regulators to refrain from impeding technological advances in deep learning while at the same time being able to clearly articulate what constitutes abusive conduct and be able to detect such conduct.