Join The Discussion | Ai4 Finance 2019
September 1-2, 2020 | MGM Grand, Las Vegas
Algorithmic trading is an automated trading system with the goal of making a profit – also known as black box trading. Typically these systems rely on complex math and high-speed computer programs to fulfil orders, and many fall under the category of high-frequency trading (HFT). HFT relies on algorithms to provide a real-time accurate prediction of the price movement of stocks. This prediction is a perfect area to apply AI.
Over time, as the adoption of these systems has increased, algorithmic trading and HFT have made major changes to the market microstructure. Since the computerization of order flow in the early 1970’s, algorithmic trading has been on the rise. Now, it’s estimated that over 75% of all trades are automated.
The problem with trading algorithms is that they’re incredibly complex. The average game of chess, for example, takes about 40 moves. The game of Go is about 200 steps long. However, a medium-frequency trading algorithm might consider a new move every second for 3600 steps per hour.
The average human-written algos are tens of thousands of lines of handwritten code, and that can become extremely unwieldy. Further complicating the matter are regulations like MiFID II and the concept of “best execution.”
At many financial institutions, machine learning – and particularly deep learning – is seen as a potential solution to the chaos. While these AI solutions can be even more opaque than their multi-thousand lined, human-written counterparts, they are, at least, less complex.
High frequency traders can apply a deep learning model to predict the stock price movement in conjunction with a financial model to place orders based on the predicted movement. In order to train the AI system, they use data from the previous day combined with the live price data. As you might imagine, the success of any AI-based algo trading system is dependent on the model, but when done right, profits can skyrocket.