For many years, financial sentiment analysis relied on simple word lists. Take for instance, on an earnings call, that would mean counting a CEO or CFO’s positive phrases like “strong growth” and negative ones like “unexpected losses”—to assign a sentiment score. This rules-based system was transparent and easy to explain.
While LLMs are more complex and costly, the study found their fine-tuned strategies could have delivered double the excess return compared to traditional approaches—particularly as market inefficiencies shrink. For example, a long-short strategy using LLM-based signals achieved 8.4% annual returns, twice the performance of traditional benchmarks (4.2%), according to S&P Global Market Intelligence.
“The real edge is precision,” said Mengmeng Ao, quantitative research analyst at S&P Global Market Intelligence. “Lexicons do well at the headline level, but LLMs separate what’s material from what’s noise. That context is what investors care about.”
Other key findings: When LLMs flagged highly important financial events, sentiment signals delivered 6.4% excess annual returns—double those for medium-importance events (3.2%) and nearly four times those for low-importance events (1.7%).
It seems that LLM-driven investment strategies have the potential to reshape how the markets move.
