Narrative Alpha
Can public investment narratives predict future returns without look-ahead bias?
Research question
Can time-stamped public investment narratives be converted into falsifiable signals without using information that was unavailable at the decision time?
Method
- Froze 1,983 bilingual posts from 10 creators, then reduced 138 raw extraction candidates to 58 high-precision claims through an audited eligibility filter.
- Constructed creator-balanced signals and aligned them with later adjusted prices, transaction costs, placebo tests, robust inference, and source/time ablations.
- Preserved publication-time provenance and hid performance metrics whenever a run contained fewer than 20 independent observations.
Result
The primary multi-source experiment produced only three independent observations; the single-source diagnostic produced six. More posts increased claims but not independent multi-source events, making source overlap and time coverage the binding constraints.
Boundary
The result is Exploratory only — not validated alpha. Historical
availability, a locked out-of-sample test, sufficient sample size, placebo
performance, and multiple-testing controls still prevent an alpha claim.