Market Beliefs about Open vs. Closed AI

Battle Count: 64

Highly relevant for event-driven strategies around AI model releases. The paper identifies a systematic, statistically significant pattern where bond yields move in opposite directions (~20-30 bps differential) around open vs. closed AI releases. This could inform fixed-income trading strategies, duration positioning, and cross-asset hedging. The differential response across treasuries, corporate bonds, TIPS, inflation expectations, USD, and equities provides multiple tradeable signals. However, the small sample and evolving nature of AI releases limit immediate practical deployment.

The methodology is standard econometrics (OLS event study with fixed effects, HAC standard errors, permutation tests). Implementation requires: (1) constructing event windows from model release dates, (2) handling overlapping windows via joint regression, (3) computing cumulative returns, (4) running permutation tests with 5000 replications. The main complexity lies in data collection and event classification rather than the statistical methods themselves.

The paper uses publicly available data (FRED bond yields, Yahoo Finance equity prices, LMArena scores, Metaculus forecasts). Model release dates are listed in Table A1. However, the specific regression code and event window construction details would need to be replicated. The methodology is well-documented with equations provided. No GitHub repository is mentioned.