THETRADETECHFX DAILY from the floor
LLMs opening up new alpha opportunities in systematic trading
How do you know when a signal has genuinely decayed versus experiencing a temporary drawdown? The key is distinguishing expected statistical variation from evidence that the underlying edge has disappeared. We continuously compare realised performance against the signal’ s historical distribution. Drawdowns, hit rates, information coefficients, and risk-adjusted returns are assessed relative to what would be expected under normal conditions.
We also examine whether the economic rationale behind the signal remains valid. If the market inefficiency the signal exploits has changed due to structural shifts, increased competition, regulation, or changes in market microstructure, that is a stronger indication of genuine decay than a poor performance streak alone. Importantly, we avoid making decisions based on a single metric or a short sample.
A signal is considered genuinely decayed only when its statistical evidence or its economic justification consistently deteriorate. This disciplined process helps avoid abandoning robust signals during inevitable drawdowns while ensuring obsolete strategies are removed in a timely manner.
Are LLMs creating new alpha opportunities or simply improving research efficiency? Traditionally, news sentiment has been measured using dictionary-based approaches that count the amount of positively and negatively connoted words. While effective to some extent, this method has important limitations.
From an investment perspective, the sentiment embedded in the outlook is far more relevant, as historical information is generally already reflected in market prices.
LLMs provide a significant advantage because they understand language in context. Rather than simply counting words, they can identify forward-looking statements, separate them from the rest of the text, and specifically evaluate the sentiment of the outlook. This results in a more precise and economically meaningful sentiment signal.
Sitting down with The TRADE, MARKUS EBNER, head of multiasset at Quoniam Asset Management, argues that AI and LLMs are opening up new alpha opportunities in systematic trading, but firms need disciplined signal monitoring, robust governance and faster processing to distinguish genuine innovation from noise and decay.
What governance concerns arise when introducing AI tools into the investment process? Introducing AI into the investment process raises governance issues like model performance, responsibility, operational resilience, and regulatory oversight.
A robust governance framework must establish clear ownership of AI tools, including who approves their use, validates outputs, and monitors ongoing performance. AI models must be subject to the same level of oversight as other investment models, with regular validation, documentation and performance reviews. Investment committees must understand where AI is used in the investment process and ensure that AIinformed decisions are always explainable.
Data governance is also important. Asset managers rely on high-quality proprietary research data, so controls are needed around data quality, licensing, confidentiality, cybersecurity, and the use of third-party AI providers. Firms must also guard against inaccurate or fabricated AI outputs and ensure that investment decisions are independently verified before implementation.
From a broader governance perspective, firms should update their risk management and compliance frameworks to address AI-specific risks, including model risk or concentration on third-party vendors, and evolving regulatory expectations.
Clear policies on acceptable AI use, staff training, record-keeping, and auditability are essential. Ultimately, strong governance enables asset managers to capture AI’ s benefits while preserving fiduciary standards and demonstrating effective oversight to regulators and clients.
Which part of the systematic FX workflow remains hardest to automate? The real challenge lies in analysing the enormous volume of news using LLMs within a limited time window. In systematic trading, speed is critical, as even small delays can reduce the value of a trading signal. While LLMs provide powerful capabilities for understanding complex financial text, processing such large datasets quickly and consistently remains demanding.
Every trading day, we collect all relevant news published during the previous 24 hours at 8am CET. The articles are then processed by our LLM-based pipeline, and the resulting signals are validated through automated quality checks before trades are generated. The entire workflow must be completed before the European stock exchanges open at 9am CET, ensuring that the strategy can react to fresh information and enter the market as early as possible.
The official newspaper of TradeTech FX 2026 TheTradeNews. com 17