Designing Stateful Prompts for AI Crypto Trading Systems
Summary
This article presents prompt design as one part of building an AI-assisted trading system. It argues that instructions should provide current market and account context, specify the desired decision format, and define the model’s role and risk constraints. It describes a two-layer setup: persistent system instructions establish the trading framework, while each user prompt supplies live status such as positions, funds, and recent performance.
The proposed refinements include requiring balanced long and short analysis, giving the model records of prior decisions, adjusting risk by instrument performance, and pausing trading after repeated losses. The article reports changes in recommendation balance and describes examples of risk adjustments, but does not provide a rigorous methodology, complete results, or controlled comparisons. It also notes that historical strengths can deteriorate when market conditions change. Specific model-generated prices and outcomes are examples from the text, not validated trading recommendations; prompt quality alone cannot replace data controls, backtesting, monitoring, and risk limits.
Key ideas
- Trading prompts should include relevant market context, account state, and a clear output format.
- A system prompt can define the model’s role and risk framework while each request supplies current conditions.
- Requiring separate long and short analyses is presented as a way to reduce directional bias.
- Trade history and cooldown rules can inform position sizing and temporarily restrict instruments after losses.
- The reported prompt improvements are illustrative, and changing market regimes can make historical performance records unreliable.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.