AI's Limitations in Trading: Insights from Quantum Signals' Yianni Gamvros

Wall Street banks have significantly invested in AI digital assistants over the past year, employing these tools primarily for research, coding, and knowledge management. While this broad adoption boosts productivity, the trading desk reveals a notable gap where AI struggles to deliver value effectively.

According to Yianni Gamvros, CEO and co-founder of Quantum Signals, the challenge lies in the nature of intraday trading, which relies heavily on rapidly changing numerical market microstructure data. A large language model excels at managing textual data, but it lacks visibility and effectiveness in interpreting real-time price and volume information crucial for timely trading decisions.

The Challenges of AI in Live Trading

During the execution of trades, understanding market movements over minutes or hours demands real-time insights into liquidity and volatility. Unfortunately, general-purpose AI tools are not designed to handle the speed and numerical precision required in these scenarios, leading to discrepancies between potential and actual performance in trading environments.

The Shortcomings of AI Trading Projects

Many AI trading initiatives falter at the transition from promising prototypes to operational systems. The initial stages involve complex offline tasks like data preprocessing and model training, which must be repeated robustly in real-time during market engagements. Gamvros points out that the true test lies in the system's ability to deliver consistent results rapidly, something that proves challenging when balancing the need for accuracy with the speed of execution.

General-Purpose vs. Finance-Native AI

While general-purpose AI excels at tasks like scrutinizing news articles and identifying macroeconomic trends, it falls short for quantitative traders focused on intraday strategies. Gamvros emphasizes that achieving reliable predictions requires innovations tailored to financial market dynamics rather than simply employing larger AI models.

The Role of Infrastructure and Expertise

For firms lacking proprietary data and infrastructure, developing a competitive AI trading capability poses significant hurdles. Gamvros underscores the importance of having a finely-tuned combination of sophisticated data management and engineering resources to exploit AI effectively. Without these foundations, firms should consider third-party solutions as a viable alternative.

Combining AI Insights with Financial Acumen

Despite AI's potential to cover a large portion of analytical capabilities—estimating that it can get you 70 to 80 percent of the way there—human expertise remains vital. Practitioners must refine AI outputs to ensure they yield actionable insights. Gamvros indicates that Quantum Signals curates prediction streams to demonstrate the technology's utility, but the human touch is essential in translating algorithmic output into real-world trading strategies.

Current AI Capabilities in Intraday Trading

AI can now effectively identify and predict key market dynamics such as price swings, liquidity, and volatility. While AI excels in pattern recognition, translating these findings into actionable trading strategies requires substantial further development. Experienced traders are also seen testing complex market behaviors beyond surface patterns, which remains a task AI has yet to master, leaving questions about when these capabilities will be fully realized.

Common Misconceptions About AI in Trading

There exists a notable disconnect in the understanding of AI's capabilities within trading circles. Many associate AI solely with text-related tasks, while overlooking its analytical potential in quantitative scenarios. Additionally, some skeptics dismiss AI due to past inefficiencies in traditional machine-learning techniques—a mindset Gamvros sees as a barrier to fully leveraging new advancements in AI technology.

Future Directions for Reliable AI Trading

For AI to achieve reliability in trading, a shift in foundational finance models is necessary. Companies like Citadel, Jane Street, and HRT are already employing foundational models specifically designed for finance, which enables them to excel. Without access to such resources, firms not adopting these advanced models risk falling behind in a highly competitive landscape.

The evolving capabilities of AI in finance suggest that the sector must continue to adapt, focusing on enhancements in model specificity, infrastructure adequacy, and data richness. Only then can firms unlock the full potential of AI in trading.