December 06, 2018
by Howard L. Kramer


Deep learning, a subset of machine learning, has begun to migrate from hedge and prop trading firms to some large bulge bracket firms who are looking to enhance their alpha returns from trading.

 A year ago, I wrote several brief articles about the impact of deep learning on securities trading and regulation.  In short, deep learning is a form of machine learning, which is a computer technology that employs algorithms to learn from data and patterns of data to produce decisions without the need to reprogram the algorithms.  Deep learning uses neural networks to enhance the ability to perform more refined calculations by adding extra layers of analysis (i.e., self-learning algorithms) between the data input and the final result.  Its benefit is a potentially more exact predictive outcome, thus providing an additional “edge” in trading.  The process is very expensive to employ because it requires far more computing power and resources than regular machine learning to attempt to obtain a marginally better result.

 At the time of my articles, some large hedge funds and proprietary hedge funds were the primary users of deep learning for trading due to the immense computing costs entailed.  I envisioned that, as the cost of computing power continued to decrease, the use of deep learning might migrate to other market participants.   From media articles over the past few months, this seems to be occurring.  Several large bulge bracket firms and banks have hired quant traders with the expertise to lead an effort to develop deep learning trading methodology at these firms and banks.  This occurrence is not surprising as these large entities have the resources necessary to accommodate the huge computing costs involved.  In addition, these entities have the resources to pay to obtain talent in this new area in an attempt to outrace their competitors.

 It remains to be seen if the bulge bracket firms can be successful in competition with some large hedge funds and prop shops in the use of deep learning for trading, and vice versa.  Deep learning is still a risky venture in terms of potential success for something as erratic as financial markets.  The recent market turbulence might just be a good test for the value and limitations of the current state of deep learning technology.

About Blockchain Law Center

Blockchain technology utilizes a distributed digital ledger to record and track information, and can be leveraged to gain transparency and certainty in transactions ranging from cryptocurrency to supply chain tracking.  This blog provides information on the legal developments surrounding implementation of blockchain technology, with an initial focus on the financial services sector.