Exploring AI Mentions in Earnings Calls and Building Thematic Portfolios

Published in

LSEG Developer Community

4 min readSep 4, 2024

You can view the whole article on the LSEG Developer Portal here.

Overview

In recent years, artificial intelligence (AI) has emerged as a central pillar of innovation across various industries. As companies increasingly emphasize AI in their business strategies, earnings calls have become an important source of insights for investors seeking to understand corporate priorities and market trends.

This article examines how the frequency and sentiment of AI mentions in earnings calls have evolved over time and whether these patterns can indicate a company’s long-term growth potential. We developed thematic portfolios based on these insights and tracked their historical performance to assess whether investing in AI-focused companies can yield significant returns.

This article is the brief overview of the developer article published on LSEG Developer’s Portal.

Setting up the dataset

The article starts with building the transcripts dataset for the constituents of Russell 1000 along with the industries they represent. We use Refinitiv Data Libraries for Python to get index constituents and the sector information and LSEG MarketPsych Transcript Analytics API to ingest all sentences from corporate earning calls mentioning word AI. The final dataset contains the following columns:

Exploring AI Mentions trends across time and sectors

In this section we provided an analysis of AI mentions in earnings calls, focusing on both temporal and sectoral trends.

Aggregated Analysis

Here we produced visualisations showing the changes in AI mentions and emotions around it over time for all constituent companies.

Sectoral Analysis

In this section we produced similar to above plots for different sectors and showed the sectoral distribution of the frequency of AI mentions in the transcripts alongside with the distribution of the specific emotion or sentiment.

Company-level Analysis

Here we analysed how the frequency and emotions of mentioning AI evolved through time for a single company. We also looked into the some of the emotions and sentiments for the top 10 companies by the number of AI mentions in the first quarter of 2024.

Constructing Thematic Portfolios based on the AI mentions

In this section, we constructed portfolios and tracked their performance over the analysis period. We employed two approaches for portfolio construction:

Clustering Approach: In this method, we divided our universe into two clusters based on the number of mentions, emotions, and sentiments.

We observed the highest cumulative returns in the weighted portfolio, followed by the higher mean portfolio. The lower mean portfolio, which includes companies with lower positive and higher negative emotions, has the lowest returns. Regarding asset allocation, the number of assets in each cluster varies significantly from quarter to quarter.

Quantile-Based Portfolios: for this approach, we categorized the companies into two quantiles based on a specific metric, e.g. the number of sentences.

With this approach the cumulative portfolio returns generally align with the clustering-based approach, with the weighted portfolio having the largest return. Additionally, both the weighted and highest return portfolios outperform the cluster-based weighted and higher mean results.

Check out the main article for more graphs and the source code.

Exploring AI Mentions in Earnings Calls and Building Thematic Portfolios

Overview

Setting up the dataset

Exploring AI Mentions trends across time and sectors

Constructing Thematic Portfolios based on the AI mentions

References

Downloads

Related Articles

Written by Haykaz Aramyan