The sentiment conductivity hypothesis: documenting lag compression in market microstructure during the LLM Era

The temporal gap between macroeconomic news release and order flow response in U.S. equity markets has compressed sharply since 2023. We advance the Sentiment Conductivity Hypothesis, which posits that Large Language Models embedded in trading algorithms accelerate the translation of sentiment in unstructured public information into quote updates and liquidity provision. This letter introduces the LLM Reaction Index (LRI), a descriptive measure of news-to-quote lag compression comparing a pre-LLM benchmark period (2018 to 2019) with a post-LLM period (2023 to 2024). Using 50 matched macroeconomic news events affecting U.S. equity ETFs, we find median lag compression for CPI releases from 450 ms to 120 ms. After accounting for estimated infrastructure speed-ups, the remaining 230 to 280 ms compression likely reflects multiple mechanisms, including non-LLM algorithmic improvement, shifts in market participant composition, and possibly LLM-based text interpretation. The design does not isolate the LLM channel from these concurrent changes. The LRI is negatively associated with post-news Quote-to-Trade Ratio (β = −0.40, p = 0.008) and positively associated with Volume-Synchronized Probability of Informed Trading (β = 0.32, p = 0.005), with R2 = 0.31. Effects are 45% stronger for linguistically complex releases, consistent with text-interpretive mechanisms. The main contribution is the LRI as a framework for tracking information-processing speed across technological regimes.