2025-02-23-美联储-骚动研究_通过电话会议记录衡量人工智能研发（英）_26页_1mb

Research in Commotion: Measuring AI Research and Development through Conference Call Transcripts
Paul E. Soto
Federal Reserve Board, 2025

Abstract

The paper introduces the AIR Index, a novel measure of corporate AI Research & Development (R&D) using natural language processing (NLP) to compare earnings conference call transcripts with academic AI research papers. The AIR Index shows varying adoption across industries—especially in manufacturing, computing infrastructure, and education—and is positively associated with market valuation (Tobin’s Q), short-term abnormal returns, and capital expenditure growth. However, it finds no significant effects on productivity or employment, indicating a gap between AI R&D discussions and tangible outcomes, consistent with the J-Curve effect and the role of upfront capital investments. The AIR Index’s impact predates the ChatGPT hype, affirming AI as a sustained technological trend accelerated since 2020.

Key Findings

1. Valuation Effects:

   • Firms with higher AIR Index scores experience immediate increases in Tobin’s Q and abnormal returns, reflecting positive market responses to substantive AI discussions (lasting ~2 years).
   • Superficial buzzwords (e.g., “AI,” “machine learning”) have minimal impact, highlighting investors’ focus on quality R&D engagement.

2. Capital Investment:

   • A one-standard-deviation rise in the AIR Index drives ~1% year-over-year capital expenditure growth for ~1 year, aligning with capital deepening.

3. Productivity & Labor:

   • Absence of substantial productivity or employment effects in the short term, consistent with prior technological innovations requiring adaptation and integration.

4. Cross-Time Analysis:

   • The AIR Index shows persistent positive associations with valuation and investment growth even before ChatGPT-era hype (~2004–2020 data), indicating that Generative AI is an acceleration of an ongoing trend.
   • High-growth industries (e.g., educational services, computing infrastructure) exhibit stronger adoption post-2022, suggesting GenAI’s role in scaling AI usage.

Conclusion

The AIR Index captures firm-level AI engagement and market valuation effects independently of superficial hype. While it correlates with capital investments, productivity gains must await organizational adaptation. This methodology underscores the need for dynamic measures to track AI integration and its economic impact.