How to Prepare for Earnings Calls Using Machine Learning Techniques

Methodology + White Paper

How to Prepare for Earnings Calls Using Machine Learning Techniques

earnings paper

The last two decades have seen an exponential increase in technological capabilities and access to data. Technological advancement has provided endless possibilities for developing insights. Within the context of capital markets, text analysis and natural language processing enable market participants to derive insights from news articles, social media, and earnings calls in addition to more structured sources such as annual filings. The inclusion of unstructured text data, primarily analysis of earnings call transcripts, in financial analysis is swiftly becoming ubiquitous across both quantitative and fundamental strategies.

In this white paper you will learn about the following takeaways:

  1. The buy-side and sell-side alike are using machine learning techniques to process text and incorporate the information in their processes
  2. The investment community focuses not only on what is said, but also on how it is said
  3. The role of the IRO now involves understanding how the market will react to corporate communications and create messaging that is consistent with the story they are trying to convey

Fill out the form to the right to access the How to Prepare for Earnings Calls Using Machine Learning Techniques White Paper >>>

 

Nasdaq Supports Issuers

Where does your company stand during Earnings?

  • Focus on ESG and Payout, be aware of emotional language

    Undervalued companies spend more time on prepared remarks than other companies with a higher valuation.
    ACCESS CASE STUDY
    Wall Street Bull statue in Manhattan

METHODOLOGY

Nasdaq Strategic Capital Intelligence Team Analysis:

  • 3,000 2019-2020 Earnings Transcripts

    - Separate prepared comments from Q&A to highlight disparity between presentation & discussion.
    - The team then ran the transcript components through both sentiment and topic programs.

  • Sentiment Analysis

    - Transcript components are tokenized and scored, both for Polarity and Subjectivity.
    - The algorithm creates an aggregate score for the entire transcript by stripping out sentences that contain no information. For example: “I will now turn it over to CFO to discuss operating results” or “We will now open it up for Q&A”.
    - Median Polarity and Subjectivity scores for information rich tokens of both the prepared text and Q&A.

  • Polarity & Subjectivity

    - Polarity is the relative spectrum between positive and negative sentiment: Polarity is scored on a scale of negative one to positive one, where a negative one would indicate overwhelmingly negative tone and a positive one would indicate entirely positive tone. The median Polarity score for S&P 500 companies over the past 6 quarters is 0.40 for the prepared remarks and 0.45 for the Q&A.
    - Subjectivity highlights the spectrum between emotion and fact: Subjectivity is scored on a scale of zero to one, where a zero would indicate completely factual information and one would indicate entirely emotional language. The median Subjectivity score for S&P 500 companies over the past 6 quarters is 0.14 for the prepared remarks and 0.19 for the Q&A.

  • Topic Analysis

    - Each transcript is processed through a topic analysis to identify top topics mentioned most frequently and categorizing each topic.
    - This process uses defined inputs including over 50 proprietary financial topic dictionaries.
    - Prepared remarks and Q&A sections are run through each dictionary and assigned a count of mentions for each topic.
    - Counts are aggregated by topic category, including: ESG, Payout, Growth, Capital Structure, Capital Efficiency, Valuation, and Other.