research Market Intelligence /marketintelligence/en/news-insights/research/natural-language-processing-part-ii-stock-selection content
Log in to other products

Login to Market Intelligence Platform

 /


Looking for more?

Contact Us

Request a Demo

You're one step closer to unlocking our suite of comprehensive and robust tools.

Fill out the form so we can connect you to the right person.

  • First Name*
  • Last Name*
  • Business Email *
  • Phone *
  • Company Name *
  • City *

* Required

In this list

Natural Language Processing – Part II: Stock Selection

Architectural Shift In 5G Core Needed To Meet IoT Demands

US Consumer IoT Market: Key Penetration Rates Creeping Toward 30% in 2023

Thomas Cook Bankruptcy Case Study – What Were The Warning Signs?

US Broadband Only Homes Surge In Q3 Ahead Of Disney+ Launch

Capital Markets
Natural Language Processing – Part II: Stock Selection

Highlights

Learn how natural language processing can be used to uncover the signal within the signal for stock selection.

Access the second part of S&P Global Market Intelligence's research report and make stock selection decisions with confidence.

Sep. 12 2018 — Astute investors have shifted their attention to explore the information content in unstructured data sets to differentiate their alpha. S&P Global Market Intelligence’s earnings call transcripts data is one such example that may offer that differentiated source of alpha.

  • Sentiment-based signals: Firms whose executives and analysts exhibited the highest positivity in sentiment during earnings calls outperformed their counterparts by 4.14% annually with significance at the 1% level. Firms with the largest year-over-year positive sentiment change and firms with the strongest positive sentiment trend outperformed their respective counterparts by 3.07% and 3.96% annually with significance at the 1% level. 
  • Behavioral-based signals: Firms whose executives provided the most transparency by using the simplest language and by presenting results with numbers outperformed their respective counterparts by 2.11% and 4.43% annually with significance at the 1% level.
  • Sentiment- and behavioral-based signals are not subsumed by commonly used alpha and risk signals. After adjustments, the signals generated excess long-short returns ranging from 1.65% to 3.64% annually with significance at the 1% level. The sentiment- and behavioral-based signals had some of the highest information ratios among all considered strategies and are lowly and negatively correlated with each other.
  • Positive language from the unscripted responses by the executives during the Q&A drove the overall predictability of the positive sentiment signal. 
  • The sentiment of CEOs has historically been more important than the sentiment of other executives. A strategy based on the sentiment of CEOs generated 3.63% per year on a long-short basis with significance at the 1% level. 
  • The aggregate sentiment of analysts historically enhanced the predictability of the 3-month FY1 EPS analyst revision signal. A strategy using the aggregate sentiment of analysts from earnings calls yielded 4.24% per year on a long-short basis with significance at the 1% level.

Natural Language Processing – Part II: Stock Selection

Download The Full Report
Learn more about Market Intelligence
Request Demo

Natural Language Processing, Part I: Primer

Learn More

Webinar Replay: Natural Language Processing - Unlocking New Frontiers In Corporate Earnings Sentiment Analysis

Learn More