Decoding Earnings Calls: How AI Tools Can Help Predict U.S. Election Results
Frank Ling
Elections draw global attention, but making accurate predictions can be challenging. Experts try to forecast results using historical and economic data, political strategy playbooks, and polls. In this article we explore a different way to assess election sentiments. By studying what company leaders say publicly, we can spot election trends sometimes before polls and news outlets do.
No polling method is 100% accurate, and all polling methods have various limitations, including the method of reaching people (phone vs. email), weighting of responses, calculation methodology, model simulation, statistical limitations, and renumeration for participation.
To address some of these limitations, LSEG developed a prediction method based on what company executives say. Using the AI sentiment analysis tool, LSEG MarketPsych Transcript Analytics (MTA) created by LSEG and MarketPsych, we can spot early signs of possible election results.
This tool uses natural language processing (NLP) and sentiment analysis to look beyond the numbers in earnings calls and catch subtle changes in how executives speak. Unlike regular financial analysis, MTA doesn't just count keywords like "revenue growth" but attempts to measure the sentiments behind these words.
Employing this novel approach: count how many times candidates' names appear in earnings call transcripts during the 20 days before an election. The data shows that in the last six U.S. presidential elections, the eventual winner's name was mentioned much more often than their opponent's. Using this same analysis for future elections may yield predictive results which may be more accurate than other existing polling methods.
The bar height represents the number of times the presidential candidate's name has been mentioned (Buzz, right axis)
2024: October 15–November 5, Trump vs. Harris
2020: October 13–November 3, Biden vs. Trump
2016: October 18–November 8, Trump vs. Clinton
2012: October 16–November 6, Obama vs. Romney
2008: October 14–November 4, Obama vs. McCain
2004: October 12–November 2, Bush vs. Kerry
As AI improves, this text analysis method may become even more accurate. In the future, we might predict results weeks before Election Day by analyzing subtle signals like keyword patterns in earnings calls. In today's AI and information age, market wisdom can sometimes be found in seemingly mundane company meeting records.
To learn more about LSEG MarketPsych tools visit our landing page where you’ll find video tutorials and more.
