Dataset

sp500-earnings-transcripts

S&P 500 earnings-call transcripts preprocessed into optimized Parquet, tens of thousands of calls (MIT). Ready for sentiment, QA and summarization research on the large-cap earnings season.

Open on Hugging Face

Data preview

Subset

Loading preview...

Dataset card

The sp500-earnings-transcripts dataset provides earnings call transcripts for S&P 500 companies from 2014 through 2024, paired with quarterly EPS and P/E valuation metrics. It is distributed as parquet files with roughly 20,681 rows on Hugging Face under an unstated license.

What this dataset contains

Each record represents one S&P 500 company's earnings call, identified by ticker, company, cik, sector, industry, headquarters, and founded fields. Time stamping is provided through year, quarter, datacqtr, datafqtr, and earnings_date columns, while the full transcript text is stored in a transcript field. The accompanying CSV transcript_coverage.csv documents which companies are included in the sample.

Structure and fields

The schema mixes string identifiers (ticker, company, sector, industry, headquarters) with numeric fundamentals including eps12mtrailing_qavg, eps12mtrailing_eoq, eps12mfwd_qavg, eps12mfwd_eoq, eps_lt, peforw_qavg, and peforw_eoq. A single train split holds 20,681 rows, delivered in parquet and optimized-parquet formats with tags indicating use for text classification, question answering, summarization, and feature extraction. The dataset is English-language and tagged for sizes between 10K and 100K rows.

How analysts use it

Quants can join the transcript text with the embedded EPS and P/E fields to test whether forward-looking language in management commentary predicts subsequent revisions to eps12mfwd_qavg or peforw_eoq. Long/short equity teams can build sector screens using sector and industry to compare tone across Health Care, Technology, or Financials buckets. NLP teams can also fine-tune summarization or question-answering models directly on the transcript column for downstream event-driven signal work.

Documentation & references 1