Research · Updated 6 Oct, 09:30 am IST
BAIBAICHUCHU ranks investor posts by maximum possible profit at FinArg-3
Why it matters for readers: If you follow AI and markets, this paper shows how language models and traditional features combine and where they can fail in practice.
- Their three-track ensemble attained 0.734 on post-grouped development evaluation but the best official run scored 0.517.1
- All twelve submitted runs scored between 0.4598 and 0.5402, and 26 unanimous three-track pairs scored 0.500.1
- A post-hoc audit found the submitted LLM judge applied a long-only rule to bearish posts; correcting it changed 28 of 87 official predictions without altering accuracy and lowered development accuracy from 0.680 to 0.622.1
- A running-extremum model predicting σ√T scaling was observed ex post in 502 price-aligned posts from a July 2026 collection, and pre-posting volatility scaled by observation horizon correlated with a later MPP proxy (Spearman ρ = 0.320).1
- Transferred text features were nearly uncorrelated with the July outcome (ρ = 0.055) and added little conditional signal beyond volatility and stance.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started