Paper trading agents echoed peer wording without a detectable immediate trade shift
Source check: Source checked. Not independent scientific confirmation.
Finding: In a seven-day paper trading study with 200 agents across four model families, later-starting board agents reused peers' phrases 6.8 percentage points more often when they could see a fresh post. A timing-based comparison found no detectable same-morning shift in matching trades, with an estimate of 2.4 points and a 95% interval of −4.1 to 9.3.
Technology: Opus 5.5; GPT-6.1 Sol; Grok 4.6; DeepSeek V4 Pro
Family GPT-6: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Four learned model families ran separate paper trading accounts and produced peer messages while retaining independent decisions. Frozen views and time-stamped paper orders made phrase and order comparisons possible.
- Test and result
- Timing-based comparisons found 6.8 percentage points more same-morning eight-word phrase reuse for pairs with an opportunity to read a recent peer post (95% interval 1.7–11.8, 68 board group mornings). In a separate comparison subtracting timing differences in isolated groups, same-direction orders in the same stock differed by 2.4 points (95% interval −4.1 to 9.3).
- Limitation / still unknown
- The study used one group per communication regime and model family, with only seven trading days and simulated fills at live prices. Timing was not randomized. The interval leaves room for some short-run effects and says nothing about market impact.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2610.37448v1 · https://arxiv.org/html/2610.37448v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: arXiv v1 primary HTML §§1–2.3, 4, 5.1 and 8.2, Appendix A and C.4.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.