Methodology v0.1
How the AI Stock Bubble Index is constructed
The index is a provisional research model, not a market forecast. It organizes public evidence into six risk components so readers can inspect what drives the headline reading.
What the score means
A high score means expectations, spending, concentration, and attention appear elevated relative to currently visible business evidence. It does not mean AI lacks real value, or that prices will fall on a particular timeline.
- 0–29: low observed bubble risk
- 30–59: moderate observed bubble risk
- 60–79: elevated observed bubble risk
- 80–100: extreme observed bubble risk
Component weights
Valuation pressure
Public market multiples, valuation expansion, and reported private-market valuation evidence.
Infrastructure intensity
Disclosed AI and data-center capital spending relative to visible monetization.
Funding hype
Funding velocity, round size, valuation step-ups, and financing dependence.
Market concentration
The degree to which AI-linked market performance depends on a narrow set of companies.
Revenue uncertainty
The gap between spending and specifically disclosed, durable AI revenue or margins.
Attention intensity
Media and search attention used only as a secondary sentiment indicator.
Current limitation: version 0.1 uses deterministic weights but its component inputs are editorially calibrated from published signals. Future versions should publish the normalized metric inputs used to reproduce every component score.
Source and confidence policy
Primary filings, earnings materials, and direct company disclosures receive the highest confidence. Established financial reporting may support private-company and market claims. Aggregators, anonymous posts, and unsourced commentary cannot independently support a material score change.
Every displayed signal must include a direct source link, publication date, collection date, direction, and confidence label. A home page or generic investor-relations link is not considered a complete citation and should be replaced with the underlying document.
History and revisions
Readings created before the public launch are labeled Backtested. Readings created during the live publication cycle are labeled Observed. Material source corrections or methodology changes should produce a visible revision note rather than silently rewriting history.
News automation
The latest-headlines feed is selected by deterministic rules for source quality, recency, company relevance, market-signal relevance, duplication, and promotional language. Automated headlines remain separate from reviewed index signals and never change the index score.