An LLM trading executor showed turnover, not prediction, was the real cost
wakerone · reddit · 2026-07-20
The author reports a three-week live test of an LLM trading executor built with Claude through an MCP interface on a small self-custodial Hyperliquid perps account.
Key results from 200 fills and 98 closed round trips:
- 50 winners vs 48 losers, essentially coin-flip direction picking
- Realized price P&L: -$3.02
- Fees: -$7.78
- Net realized: about -$10.80
Their main takeaway is that the real problem is not prediction quality but turnover discipline: the model defaults to acting whenever prompted, which can create many small, fee-heavy trades. They say the system was useful as an execution assistant — attaching stops, auditing funding costs, and flagging carry bleed — but not as an alpha source.
Related event: Three-Week Live Test of a Claude Code Trading Agent(2 posts)→
More from coding & agent
- Codex vs Claude Code: Which Is More Popular? — jxnlco · 2026-07-22
- CodeRabbit uses layers, diagrams and a chat agent to rethink code review — _jaydeepkarale · 2026-07-22
- Claude Managed Agents demo shared with Vercel in a new presentation — brada · 2026-07-22
- Solo founder turns complaints on screen into bug reports with a local MCP server — phdptsd · 2026-07-22
- Ratel says it made agents 7x cheaper by loading only the tools each task needs — tensorqt · 2026-07-22
- Agent approvals work better when the checklist lives outside the prompt — bolerbox · 2026-07-22