Misconceptions About A.I. and Securities Trading
A slew of articles has appeared recently about the growing use of artificial intelligence (“AI”) and neural networks for securities trading. Most of these articles conflate basic algorithm refinement with deep learning and neural networks. As I noted in an article I wrote last December, I am confident that some forms of deep learning are being used by a few hedge funds and proprietary traders, and that such use will grow quickly over time. The notion, however, that it is taking hold on a widespread basis seems grounded on a misunderstanding of the nature of deep learning.
AI is a concept that is broadly applied. In general, AI involves the development of intelligent systems not produced by nature. Machine learning is a subset of AI where computers can learn automatically from past data and improve programs on their own. In my article last December I explained in basic terms the difference between machine learning generally and a subset called deep learning. The latter uses extra layers of analysis between data input and the result produced. Deep learning requires enormous computing power to process multiple layers of inputs and outputs to obtain a marginally better predictive ability. For most securities trading entities, the expense to build and maintain that level of computing power is not economically viable at present.
Two trends will inevitably change this situation. First, the cost of computing power, and thus the cost of implementing machine learning and deep learning, continues to decrease at an expeditious rate. Second, as the edge from current forms of high frequency trading declines further, proprietary trading entities will seek other means of gaining superior performance results. The entities who can successfully employ deep learning before others will obtain that sought-after edge.
Many commentators will have you believe that we have already reached this point and that entities are racing to add new forms of AI functionality to their trading. By itself, that does not convey much information as it does not address whether entities are using deep learning. At present I believe that some hedge funds or trading entities are employing deep learning. I have viewed the websites of a few entities purporting to incorporate such programming technique into their trading. At present, however, I believe that such entities are the exception, and are not widespread. My belief was reinforced in recent discussions on the topic with several securities and derivatives traders and market makers who told me that their firms are not prepared now to make the type of large expenditure necessary for deep learning. For sure, many trading firms are using “big data” and some are employing machine learning to enable their trading algorithms to learn from patterns of copious amounts of data without the need to reprogram the algorithms. By itself, that is not much of a revelation. What will be even more informative for securities regulation is when a greater number of trading entities begin to employ deep learning to underpin their trading strategies.
- Two Murphy & McGonigle Lawyers Recognized as “Rising Stars” by Super Lawyers New York
- 19 Practices Honored and Six Earn National Rankings in Best Law Firms 2020 Survey
- “Post Lava Jato World” Event at Murphy & McGonigle to Feature Leading Authorities Involved in the Enforcement Effort
- Murphy & McGonigle Diverse Lawyers Participate in Annual MCCA Conference
- “Commodity” is a Defined Term!