Building a Single-Asset Trading Loop That Updates Its Playbook
Summary
This article describes a single-asset trading system that cycles through market sensing, decision-making, execution, trade review, and playbook updates. It structures technical indicators and cached news for an AI model, which returns a directional signal and confidence. A rule blocks new or larger positions when technical and news views conflict. Execution parameters can come from the current playbook, while code-enforced bounds constrain position size, stops, leverage, and additions. After a full trade closes, the system records an AI-generated review; later updates use accumulated reviews and performance summaries to revise rules and selected parameters.
The author clarifies that this is not online model training: the model weights remain unchanged, while the external playbook and bounded parameters evolve. The article gives a design rationale and example safeguards, but no evidence that the loop improves trading performance. It highlights weak early evidence, noisy single-asset samples, subjective AI attribution, and the absence of robust out-of-sample validation or drift detection. It recommends observation or notification mode before considering live trading.
Key ideas
- The system cycles through structured market inputs, AI decisions, trade execution, reviews, and playbook revisions.
- Technical and news signals must not conflict before the system can open or add to a position.
- AI-suggested parameters are clipped to human-defined limits, while hard risk controls remain in code.
- The playbook changes through stored reviews and parameter suggestions rather than updates to model weights.
- Small samples and retrospective AI explanations can mistake noise for durable trading lessons.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.