The Money Makers are Embracing Machine Learning

John Murray
Primalbase
5 min readJul 30, 2019

--

Buy low. Sell high. That’s the underlying goal for any kind of investment, but navigating the financial markets is far more complex than this and involves a great deal of volatility. The big firms dealing with hedge funds and multi-billion dollar portfolios are always looking for new ways to stay ahead of the curve when it comes to analysing market trends and developing strategies that will see reliable returns for their clients.

This is exactly why machine learning has attracted so much attention and investment in the financial trading world. Firms including JPMorgan and Morgan Stanley have become embroiled in a high-tech arms race in recent years, pouring billions of dollars into incorporating machine learning platforms into their infrastructures and hiring developers and researchers into dedicated data science equity teams.

How widespread is the use of machine learning within the hedge fund investment market, and how much of an impact has it already made?

Investing is a Data Goldmine

Machine learning is built upon a solid bedrock of high quality, high volume data. The financial industry already utilises it across the board, and has seen the evolution of different types of funds and specialised traders emerge to exploit it. A quant fund is a type of investment fund that selects specific securities through quantitative analysis, by means of specialised traders, often referred to as ‘quants’ themselves, building complex software models. High-frequency trading applications are typically coded in C++, while offline models can utilise MATLAB and SAS.

Photo by Markus Spiske on Unsplash

Quant funds have already established themselves as a significant avenue of investment strategy within financial markets, with some estimates in 2017 reporting quant fund managers as being responsible for 27% of all US stock trades.

Big Banks are Going Further

Machine learning and AI development has been the next logical step for massive financial institutions looking for more effective and future-proofed ways of exploiting their data streams. The goal of these technologies is to allow hedge fund managers to find trading signals based on historical data, with the minimum level of human intervention possible.

The concept is an appealing one and hedge fund managers have been quick to adopt the technology to aid their own trading processes and portfolio diversification. A 2018 BarclayHedge survey found that more than two-thirds of respondents said that they used AI/ML in some capacity to inform investment decisions and optimise their portfolios, while more than a quarter have used the technologies to automate their trading.

JPMorgan and Morgan Stanley have been the industry leaders in the aforementioned AI arms race, but several other institutions are also pouring investment into the arena. Bridgewater Associates is one such hedge fund giant that is focusing heavily on its machine learning infrastructure, as well as Simplex Asset Management in Tokyo, representing notable interest from the Asian markets.

The most recent development in the field is JPMorgan’s strategy to invest in established and emerging machine-learning statistical-arbitrage hedge funds. This suggests not only a plan to secure their own machine learning infrastructure but also speculation that this emerging section of the financial market is ripe for exploration and early-stage investment.

A Shaky Start

Despite the relatively enthusiastic adoption of machine learning in hedge fund management, it has not been an entirely smooth roll-out. There have only been modest returns overall from the strategies developed by these machine learning methods, such as the 1.1% annualised return in three years from the Man AHL Dimension fund, compared with an almost 5% gain for the average hedge fund.

Machine learning integration into hedge fund trading, like its integration into any industry, takes time, money and highly specialised expertise. This expertise is in high demand across various industries, making the reliability of forming sufficient teams to push forward with suitable machine learning development hard to guarantee. For example, a quant unit of Man Group, Man AHL, needed three years of work to gain enough confidence in its machine learning technology in order to finally dedicate client money to it.

Always Expect the Unexpected

A large degree of trepidation remains with the pairing of machine learning and hedge fund investment. In order to successfully train machine learning models, there must be a degree of hand-holding by developers, with data streams carefully vetted and whittled down into their most tightly concentrated, relevant forms. However, financial markets can be severely affected by sudden, unpredictable events and revelations, so shouldn’t any machine learning model being integrated into hedge funds be fed the biggest amount of historical investment market data possible?

Unfortunately, doing so leaves the door open to these models finding patterns that are ultimately meaningless to hedge fund managers who require far more focused prediction parameters. And of course, algorithms still have trouble with bombshell events such as random terror attacks and political events such as Brexit.

Accountability is Key

Machine learning algorithms and artificial intelligence programs are often lambasted for a lack of transparency, thanks to the ‘black box’ nature of their development and functionality. In the world of financial trading, where billions of dollars must be accounted for, any monumental slip-ups and losses require a direct and easily traceable path of culpability.

The first-ever legal case concerning the liability of AI platforms in poor financial investments is being raised in London. Hong Kong real estate mogul Samathur Li Kin-kan is suing Raffaele Costa, CEO and founder of Tyndaris Investments, over the latter’s use of a supercomputer named K1 in a robot hedge fund. Li agreed to let the computer’s AI manage $2.5 billion of investments, but it began making heavy losses, including one particularly bad day where $20 million was wiped off the portfolio.

Photo by Jp Valery on Unsplash

Such high profile cases involving AI liability were bound to happen sooner or later. It’s essential to look past the hype that often comes with machine learning integration into various industrial sectors, seeing instead its development as complementing existing skill sets rather than fully replacing them. Hedge fund managers are seeing machine learning tools not as a magic wand for great investments, but as a powerful tool which must be shaped in accordance with the confines of the markets.

--

--

John Murray
Primalbase

Senior Editor at Binary District, focusing on machine learning, AI, quantum computing, cybersecurity, IoT