Transcript AI Exposure
In plain terms
Bigram score against an AI/LLM lexicon on transcripts. High-AI-exposure firms outperformed sharply after ChatGPT launched.
How it works
Eisfeldt-Schubert-Zhang (2023) NBER w31222 "Generative AI and Firm Values": HHVT-style bigram score against an AI/LLM training corpus (ChatGPT-era ML papers + tech-press articles). Documented a 5-month L/S of ~9% post Nov-2022 on the AI-exposure decile — the cleanest paper-published AI-narrative trade in the literature. Signal is own-history z of per-call AI-bigram density: long when z ≥ +1 (firm just elevated AI narrative), short when z ≤ −1 (firm de-emphasized AI).
Data dependencies
- Daily prices
Adjusted-close OHLCV for every US-listed ticker; primary price feed.
- Earnings call transcripts
Full earnings-call transcripts (prepared + Q&A), tokenised.
Expected edge
- Reported return
- ~9% 5-month L/S post Nov-2022
- Tested over
- 2017-2023 (Eisfeldt-Schubert-Zhang)
Documented a 5-month L/S of ~9% post Nov-2022 on the AI-exposure decile — the cleanest paper-published AI-narrative trade in the literature.
Explore Transcript AI Exposure on alphactor.ai
See which tickers this family is currently firing on, with live signals and rankings.