Dataset

sp500_earnings_transcripts

Earnings-call transcripts for S&P 500 companies, tens of thousands of calls in English with timestamps, in Parquet (MIT). Direct input for call-tone and question-and-answer sentiment analysis across the large-cap earnings cycle.

Open on Hugging Face

Data preview

Subset

Loading preview...

Dataset card

The kurry/sp500_earnings_transcripts dataset on Hugging Face aggregates earnings call transcripts for S&P 500 constituents and other US large-caps from 2005 through 2025. It is distributed as a single train split of 33,362 parquet records and is intended for financial text analysis and NLP research.

What this dataset contains

The dataset holds roughly 33,000 earnings call transcripts covering S&P 500 companies and major US large-caps across all 11 GICS sectors, with quarterly coverage from 2005 to 2025. Each row corresponds to one company-quarter call and carries identifying fields such as symbol, company_name, company_id, quarter, year, and date alongside the call text. The publisher flags the corpus as suitable for text generation, text classification, and market sentiment work.

Structure and fields

Records are stored in parquet with a tabular-plus-text schema. Core string columns include symbol, company_name, date, and content (the full transcript), while integer columns quarter and year support temporal slicing, and a float company_id links to an external entity reference. A nested structured_content field breaks each call into speaker-attributed segments, enabling separation of operator remarks, prepared management commentary, and the Q&A portion for finer-grained analysis.

How analysts use it

Quantitative investors can mine the content and structured_content fields to build NLP features such as management tone, guidance language, or analyst-question counts, then join them by symbol, quarter, and year to factor exposures or event-study reactions. Because the corpus spans two decades of US large-cap calls, it also supports backtesting of textual signals across market regimes. Researchers typically load it via the datasets library and filter by symbol and year, for example isolating Apple transcripts from 2022.

Documentation & references 1