← News

Agents & MCPSources: arXiv · 16 Nov 2026

FinPanel picks trader personas per stock before any coordination begins

Researchers propose task-conditioned persona selection for LLM trading agents, arguing the cast of agents should be chosen from each task's characteristics rather than reused across every task.

In plain words

Large language model trading systems usually use the same fixed set of agent roles for every task. FinPanel instead has a recruiter choose a panel of named trader personas for each stock, based on a statistical sketch of how that stock trades. Each persona then gives an independent daily call, and a manager commits the trade after reading their arguments.

What was announced

The authors, publishing on arXiv, argue that multi-agent LLM systems are typically built by fixing a cast of roles and designing how they coordinate, leaving the question of which agents should serve a given task to the designer. They instantiate task-conditioned persona selection in FinPanel: a recruiter selects a panel of trader personas for each stock from a direction-blind statistical sketch of its trading behaviour, each persona gives an independent daily call in its own style, and a manager commits the trade after reading the panel's arguments and desk-kept hierarchical track records. Because nothing is returned to the traders, the panel can trade a new stock from an empty record.

On eight US stocks over 2025, described as the out-of-knowledge year, the selected panel reportedly outperformed a fixed panel of the same size and three prior LLM trading agents re-implemented in the same framework, plus rule-based, predictive and reinforcement-learning baselines, with a shallower drawdown than buy-and-hold on every stock. The paper reports the panel keeps its performance with no warm-up and no record, while memory-augmented agents' returns swing with their warm-up, and that adding seats or manager reading does not substitute for selecting the right personas.

What it does not show

  • Evidence is limited to the paper's abstract; the full article was not reviewed, and the reported results are the authors' own claims without independent confirmation.
  • The evaluation covers eight US stocks over 2025, so performance on other markets, periods or instruments is not established by this evidence.

For security and IT teams

Scope
The source is an arXiv abstract describing a research paper; findings are the authors' reported results, not independently verified deployments.

Sources

Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 17 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.