• Yasir Aheer

Trading in the World of AI

by Yasir Aheer, Varun Rao, and Rahul Rao


  • Investment management and trading is an information based industry, and lends itself naturally to modern technologies such as Big Data and Machine Learning.

  • Machine Learning (ML), a form of Artificial Intelligence, uses sophisticated software to analyse large volumes of unstructured data to find hitherto unknown patterns and predictions, assisting with investment decisions.

  • ML can have limitations and could be exposed to biases inherent in the “training data”.

Public opinion can be highly polarised when it comes to Artificial Intelligence (AI). At one end of the spectrum, AI is hailed as the next leap forward, while at the other end, as certain doom with eventual extinction of the human race. This divergent view is also reflected in various popular culture movies such as Moon (2009) on the one hand, with GERTY as man’s trusted companion, and The Matrix Trilogy (1999 to 2003) on the other with humans locked in battle with the machines for survival.

This got us thinking - with machines playing such a fundamental role in our modern society, is there any industry where we can see both ends of this divergent experience all at once? The world of trading comes close in some respects. It is a high-stakes environment with real winners and losers in terms of yields & returns and, like in most other industries, machines have taken on increasingly greater roles over recent decades.

In some respects investment management and trading require predicting the future. How will the geo-political landscape influence investor sentiment? Which companies are going to do well? Is there an economic slow-down on the horizon? Investment management is about having a perspective on the events to come and positioning your portfolio accordingly. Some technology driven approaches that have had the greatest impact on investment management and trading in the modern age are: High Frequency Trading (HFT), Algorithmic Trading and Machine Learning.

High Frequency Trading

One estimate puts HFT at 50 percent of the total stock trading volume in the U.S. High Frequency Trading became popular as exchanges offered incentives to add liquidity to the market. As an example, NYSE pays a fee or a rebate to companies that provide such liquidity. One such group on the NYSE is Supplemental Liquidity Providers (SLP) who use high-speed computers and sophisticated algorithms to push high-volume on exchanges for liquidity.

High Frequency Trading realises value through a simple concept, “the early bird gets the worm”, or as its more technical definition: Latency Arbitrage. Latency arbitrage simply means having an edge by being faster. Over the past decade brokerage houses have pushed the limits to gain a speed advantage by buying faster computing power (both software or hardware) or positioning their servers closer to the exchange computers. Some exchanges even rent space to co-locate brokerage and exchange servers. However, the benefits of pure arbitrage based on speed might be eroding as the price of maintaining that edge starts to come in line with the value delivered resulting in diminishing returns.

HFT can be a contentious topic, as it has been used for predatory strategies such as front-running, eroding returns of the “small guy”. HFT isn’t without its pitfalls either, as it is vulnerable to manipulation techniques such as spoofing and artificially induced flash crashes, such as the one in May 2010.

Algorithmic Trading

Algorithmic trading, also known as Algo-trading, uses a more traditional form of computer software. It uses computer programs that follow predefined instructions and executes them systematically at a scale and speed impossible for a human trader. An example of a pre-defined instruction could be:

  • Buy 50 shares of a stock when its 50-day moving average goes above the 200-day moving average. (A moving average is an average of past data points that smooths out day-to-day price fluctuations and thereby identifies trends.)

  • Sell shares of the stock when its 50-day moving average goes below the 200-day moving average.” [1]

Algo-trading requires little to no human intervention once the algorithm has been initiated. The algorithm continues to monitor the market and automatically executes the pre-defined order volumes once the programmed criteria are met, such as the simplified example above. Oftentimes algo-trading is used in conjunction with HFT and hence it is also exposed to the same limitations or vulnerabilities as mentioned earlier for HFT.

Machine Learning

Machine Learning together with Big Data, could be one of the biggest recent game changers for the investment management and trading industry. While HFT and algo-trading follow precise pre-programmed instructions, Machine Learning (ML), a form of Artificial Intelligence, uses sophisticated software to analyse large volumes of unstructured data to find hitherto unknown patterns and predictions, assisting with investment decisions. These software packages also have the ability to improve their performance through iterative feedback loops as more data is made available.

More recently, various AI techniques have been utilised for unstructured formats unlocking valuable “alternate data'' that can be brought to bear for investment decisions. Some forms of “alternate data” include satellite imagery, news feed, social media sentiment and geo-spatial information. As an example, GPS location data from mobile phones can help build a richer picture of foot traffic at various retail stores. Leveraging satellite photographs, ML algorithms can predict agricultural crop yields while still in the fields or sales at specific shopping malls based on the number of cars in the parking lots [2]. JPMorgan categorises data sources into three types:

A word of caution, Machine learning is subject to some important limitations. Firstly, being a specialised field there can be a dearth of expert talent. Secondly, the ML algorithms may themselves exhibit biases driven by the data used in the training process. Thirdly, current ML approaches train their models using data from the past, which inherently means that they may not be appropriate at predicting “Black Swan” events in the future. Lastly, ML models can be black boxes - it may be hard to understand the drivers behind the recommendation or outcome. Therefore, some form of human judgement is required as part of the investment process.

Medallion Fund

Any exploration of quantitative analysis and the use of modern technology in investment management or trading would not be complete without mentioning the stellar performance by Renaissance’s famed Medallion fund. The secretive firm established by the cold war codebreaker and mathematician, Jims Simmons, is believed to use quantitative analysis and modern technologies such as algo-trading and machine learning to inform their investment decisions. While the markets have been in dis-array in recent months driven by COVID-19 pandemic, the $10b Medallion Fund has out-performed its peers and delivered 24% gains year-to-date. Over the entirety of its existence from 1988 to 2018, Renaissance’s Medallion Fund has delivered 66% annualized returns (before fees) and 39% annualized returns (net of fees) from 1988 to 2018, as noted by Greg Zuckerman in The Man Who Solved the Market.


So, what is it like making investment decisions or being a trader in the modern digital world? Continuing the popular culture metaphor we started with, the experience can be summed up by two analogies. On one end of the spectrum, it is like Captain Kirk from Star Trek venturing into the unknown with trusted Mr. Data (machine learning) on his side, helping him make informed decisions. At the other end it resembles John Connor from the Terminator series battling the robots (HFTs, algo-trading) for higher yield and returns. And like most things in life, reality might be somewhere in the middle.


Disclaimer: This article is based on our personal opinion and does not reflect or represent the views of any organisation that we might be associated with.