> ## Content Index
> Fetch the complete content index at: https://fsc.stevens.edu/llms.txt
> Use this file to discover other available public pages before exploring further.

# Multimodal Behavior Analytics From Earnings Calls
- URL: https://fsc.stevens.edu/multimodal-behavior-analytics-from-earnings-calls/
- Published: 2026-11-20T15:23:55.000Z
- Updated: 2026-11-20T15:23:55.000Z
- Author: Malavika Loka
- Tags: 2026 Fall, FE 800 Projects

**Researcher:** Malea McMurray 

## Introduction

An earnings call reveals both what executives say and how they deliver it. This project examines whether features from transcripts and audio are associated with stock returns or volatility in the three days after an earnings call. 

### Methodology

The analysis compares transcript sentiment, measured as text polarity, with voice features such as the share of pitch spikes and variation in pitch. Across a dataset covering 503 companies, it examines how those features and the study’s composite score, labeled BCS in the figure, relate to three-day abnormal returns and volatility. 

### Results & Analysis

**The composite score showed little linear association with three-day abnormal returns or volatility in the full sample.**Its correlations were −0.06 with abnormal returns and 0.03 with volatility. Text polarity and the share of pitch spikes also had a near-zero correlation, indicating that the two measures captured different aspects of the calls in this dataset. 

[ ![Four panels comparing text polarity with pitch spikes, the BCS score with three-day abnormal returns and volatility, and pairwise correlations across the 503-company dataset](https://fsc.stevens.edu/content/images/2026/11/McMurray_4Panel.jpg) ](https://fsc.stevens.edu/content/images/2026/11/McMurray%5F4Panel.jpg) 

Transcript sentiment, voice features, the composite score, and three-day market reactions across 503 companies. Select the image to read the panel labels and correlation values at full size. 

### Conclusion

Transcript and audio features offer two ways to characterize how executives communicate during earnings calls. In the full sample, the composite score had only a weak linear relationship with the short-term market outcomes shown.