Search earnings calls

Use the Risk Tool to search earnings-call transcripts for a user-defined topic and generate call-level measures.

Prepare:

  • A draft or validated query.
  • The date range for the search.
  • A clear target sample for later interpretation.

The current documented workflow centers on English earnings-call transcripts.

  1. Open the Risk Tool.
  2. Enter the query using keywords, phrases, or supported query syntax.
  3. Select the date range.
  4. Choose additional search options where they matter for the research design.
  5. Run the search.
  6. Review the results overview.
  7. Open matched sentences in the Snippet Tool if the results need validation.
  8. Download the CSV output when the query is ready for downstream work.

The Risk Tool search form currently does not let users specify particular industries or firms before running a search. You can focus on firms or industries after search through output files and supported Snippet Tool filters.

Search options

In addition to the query and date range, the Risk Tool supports options for narrowing where matches can occur:

OptionUse it to
Section of earnings callSearch only selected parts of a call, such as presentation or Q&A.
Speaker affiliationFocus on corporate executives or external participants.
Include adjacent sentencesInclude one sentence before and after the matched sentence by the same speaker where available.
Search nameGive the search a label that is easier to find later.

The search engine generally operates at the sentence level. When adjacent sentences are included, the search can operate on snippets of three adjacent sentences by the same speaker.

Output

Risk Tool output is call-level. The documented metric variables are:

  • exposure
  • risk
  • positive
  • negative
  • sentiment

CSV exports include all earnings calls in the selected date range, including rows with zero metric values.

Available export concepts include:

  • Topic exposure, risk, and sentiment for the user's query.
  • Overall risk and sentiment counts that are not restricted to the user's query.
  • GVKey crosswalk output for downstream Compustat joins.
  • Search metadata for audit and reproducibility.

Authoritative file names, column ordering, and full schemas are not inferred here. Check the actual export and keep search metadata with downloaded files.

Search results are only as good as the query. Inspect matched sentences, check coverage and variation, and refine the query before treating the output as a measure.

Was this page helpful?