Abstract
As of 2023, Artificial Intelligence (AI) is playing an increasingly important role in financial trading. AI manifests in various forms, from advanced algorithmic trading systems (Algotrading) to sophisticated machine learning models, all aimed at optimizing trading strategies. The competition in this space is fierce, with both AI models and human traders striving to outdo one another, often with significant financial stakes involved.
While confidence in AI-driven strategies is crucial, success is largely dependent on data-driven decision-making, sound reasoning, and effective risk management, these days sound reasoning is not an area of AI excellence, human input is vital so extreme arrogance helps a lot 🙂 however, decision making that is strictly dependent on data is a core competency of AI.
Pegasus is an algorithmic trading (Algotrading) system I developed primarily in Python. It’s an ongoing project, reflecting the continuous nature of AI development. Unlike systems focused on natural language processing, such as some projects by Google Brain, Pegasus is designed to process financial market data using neural network algorithms. Specifically, it employs an encoder-decoder structure from the Transformer architecture, which is adapted to analyze trading data, capturing context and patterns through its neural network layers.
The system embodies my 35 years of trading experience, integrated through a fuzzy logic controller that effectively manages market uncertainties. This is further enhanced by advanced information processing capabilities and robust mathematical models, enabling sophisticated analysis and decision-making in the trading environment.
More details on actual algorithmic trading will follow soon. First, we’ll delve into neural networks and explore the Transformer architecture. For those eager to get a head start, I recommend reviewing the seminal paper on the Transformer architecture https://arxiv.org/pdf/1706.03762.pdf
If you’re looking for a more specialized and detailed discussion, this https://arxiv.org/pdf/2208.08300.pdf may be of interest.