LO
CapabilityCognitive automation2026-04-06
Top models produced valid trading-system code more than 91.7% of the time on SysTradeBench, while its authors concluded human oversight remains essential
Quantitative analystoccupation page →
Event date / reported
2026-04-06
Evidence stage
CapabilityA demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.
Tasks this bears on
Backtesting and quantitative code
Writing and maintaining the code for models and backtests, and computing performance and risk metrics correctly.
Being augmented✓ Evidence-backed
Where this applies
A benchmark of 17 models across 12 strategies that turns strategy specifications into trading-system code and iterates on it. The abstract reports that top models achieve validity above 91.7 percent with strong aggregate scores, that iteration also induces code convergence, and that human oversight remains essential for critical strategies requiring solution diversity and ensemble robustness. It is a benchmark, not production work.
What this means
Turning a written strategy into working code is largely within reach of models; what their own evaluators keep for people is judging which strategies to trust.
What it does not yet show
A benchmark; it does not measure real research work.
What you can check
Open arXiv 2604.04812 (SysTradeBench) and find "validity above 91.7 percent".
Does it change the assessment?
No — and this stage does not move it either. A "Capability" record is real evidence, but it does not upgrade a task judgement on its own. The 1 linked judgement above stand where they were.
Source
Cao, Zhang, Keung et al. — SysTradeBench: An Iterative Build-Test-Patch Benchmark for Strategy-to-Code Trading Systems with Drift-Aware Diagnostics, arXiv 2604.04812 (submitted 6 Apr 2026) · verified 2026-09-29 · Claude (VOLO agent) · interpreted 2026-09-29 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.