AI trading agent backtest: 1,155 decisions in 77 days for $0.09, still trails buy-and-hold
nikola_mr64990 · x · 2026-09-23
A developer ran a full AI trading agent backtest: 77 days, 1,155 decisions across 3 strategies (momentum, contrarian, macro) and 5 assets (BTC/ETH/SPY/QQQ/XAU) for a total cost of $0.09 (Ling-3.0-flash-VL is free on OpenRouter), completing in 87 seconds with 1.1s per call and zero failures.
Workflow: daily investment digests were fed one at a time to Jev 1.13 via OpenRouter along with current holdings; the model output buy/sell/hold probabilities and confidence, and the program executed against real hourly Binance candles, computed NAV and reconciled.
Results were promising but not magic: momentum returned +6.71% vs +13.2% for equal-weight buy-and-hold, while the other two strategies returned −1.0% and −10.3%. The bigger win was presentation: Ling-3.0-flash-VL turned a design mock into a dynamic dashboard (NAV curve, position bars, decision feed) from a screenshot alone, then playwright frame-by-frame screenshots (390 frames) were compressed into a 13-second video summarizing 95 days of news and market action.
More from coding & agent
- Open-source GUI agents top out at 8% task success on composite cross-device tasks — maier_ak · 2026-09-23
- JarvisGUI benchmark tests GUI agents across Android, Windows and Ubuntu in one workflow — maier_ak · 2026-09-23
- Cross-DURIAN: A New Benchmark for Multi-Device GUI Agents — maier_ak · 2026-09-23
- "Proactive" personal agents: reminder, proposal, or acting without asking? — sujingshen · 2026-09-23
- AWS Open-Sources Strands Harness, Claims 28% Fewer Agent Tokens — shashib · 2026-09-23
- Before your AI agent pays for you: four questions about authorization and billing — sujingshen · 2026-09-23