Expert: Agentic Bots Are Now 'Research Assistants'
According to quantitative finance expert Irene Aldridge, trading bots are evolving into agentic systems that act as "research assistants." These new bots use LLMs and reinforcement learning not just to execute trades, but to automate alpha signal discovery, scrape alternative data, and propose new strategies to human quants.
- Agentic systems in finance often employ a multi-agent structure, mirroring roles in a real-world trading firm with specialized bots for fundamental, technical, and sentiment analysis, alongside others for risk management and trade execution. This division of labor allows for a more robust and scalable approach to market analysis. - Open-source frameworks like LangChain and AutoGPT are being used to build custom agentic pipelines for quantitative finance. There are also finance-specific open-source projects, such as FinGPT, which aim to democratize access to advanced financial AI. The GitHub repository "AgenticTrading" provides an open-source orchestration framework for financial agents, demonstrating a move towards more transparent and collaborative development in this space. - Beyond executing trades, agentic AI is being developed to automate the entire research workflow, from forming a hypothesis and sourcing data to running backtests and evaluating performance. This capability aims to significantly accelerate the discovery of new alpha signals and trading strategies. - The use of reinforcement learning is a key component in these advanced bots, allowing them to learn and adapt their strategies over time based on market feedback and the outcomes of their decisions. Some models, like the AlphaQCM method, are specifically designed to handle the non-stationary and reward-sparse nature of financial markets. - Alternative data, gathered through web scraping of sources like news articles, social media, and even satellite imagery, serves as a crucial input for these agentic systems. Over 60% of hedge funds now incorporate social media data into their strategies, and the overall alternative data market was valued at $7.2 billion in 2023. - Large Quantitative Models (LQMs), distinct from LLMs, are being developed to simulate billions of potential market scenarios in real-time, offering more dynamic risk assessment and portfolio optimization than traditional models that rely on historical data. This allows for more precise decision-making by anticipating market volatility and identifying new opportunities. - The shift towards agentic AI is changing the role of the quantitative analyst from a hands-on coder to more of an "AI shepherd" or system designer. The focus is moving towards high-level strategy, validation of AI-generated models, and possessing a deep understanding of machine learning frameworks like TensorFlow and PyTorch. - As these autonomous agents become more prevalent, there are growing concerns among regulators who are currently not equipped to police AI-driven trading. Experts like Irene Aldridge suggest that regulatory bodies will need to develop their own purpose-built AI systems to effectively monitor and manage the new market dynamics created by these technologies.
