How AI is changing earnings call analysis—and stock picks

Core Insights - Analysts and investors are increasingly utilizing generative AI, particularly large language models (LLMs), for stock selection and reviewing earnings calls [1][6] Financial Sentiment Analysis - Traditional financial sentiment analysis relied on simple word lists to assign sentiment scores based on positive and negative phrases from earnings calls [2] - LLMs offer a more nuanced understanding by interpreting context and language structure, allowing for the recognition of positive sentiments even in the presence of negative words [3] Performance Comparison - LLMs have demonstrated the ability to extract insights from earnings call transcripts and convert them into actionable trading signals, matching the effectiveness of traditional sentiment models [3] - A long-short strategy utilizing LLM-based signals achieved annual returns of 8.4%, which is double the performance of traditional benchmarks at 4.2% [4] Precision and Importance of Events - The precision of LLMs allows them to differentiate between material information and noise, providing a significant advantage over traditional lexicon-based methods [5] - When LLMs identified highly important financial events, sentiment signals yielded 6.4% excess annual returns, compared to 3.2% for medium-importance events and 1.7% for low-importance events [5]