Earnings, estimates and analyst ratings

Ten years of US earnings surprises — actual EPS and revenue versus consensus estimate at the time — for every USD-reporting US-listed issuer FMP indexes. 313,626 reported earnings events as a single gzipped CSV. Computed surprise columns (epssurprise, epssurprisepct, revenuesurprise, revenuesurprisepct) are derived in the build for convenience. Read with pandas.readcsv(path, compression='gzip', parsedates=['reportdate', 'lastupdated']) or duckdb.readcsv('earnings2016_2025.csv.gz'). Foundational dataset for: post-earnings-announcement-drift (PEAD) studies, beat / miss base-rate research, earnings-quality scoring, surprise-magnitude factor construction, and tracking which sectors / size buckets systematically beat or disappoint.

Ten years of sell-side consensus quarterly estimates for 8,480 US-listed issuers, covering 175,834 quarter-periods. Each row carries low / avg / high consensus values across six fundamental lines — revenue, EBITDA, EBIT, net income, EPS, SG&A — plus analyst-coverage counts. Read with pandas.readcsv(path, compression='gzip', parsedates=['date']). Useful for: pairing with the Earnings Surprises listing for analyst-accuracy / drift research, tracking estimate revisions over time, building consensus-bias factors, training models on the dispersion (high − low) as a proxy for fundamental uncertainty.

Five years of full-text earnings call transcripts from the top ~1,000 US public companies by reporting frequency (plus Tesla and Meta, which IPO'd more recently), covering Q1 2021 through Q4 2025. 19,677 transcripts across 1002 symbols, packaged as 10 gzipped JSONL files — one file per half-year (H1 = Q1+Q2, H2 = Q3+Q4) to stay under platform per-file size limits. Each line is one transcript with fields: symbol, fiscalyear, fiscalquarter, date, content (full prepared remarks + Q&A, speaker labels preserved). Sourced from Financial Modeling Prep. Read natively with pandas.readjson(path, lines=True, compression='gzip') or duckdb.readjson('file.jsonl.gz', lines=true) — concatenate all files for the full 5-year corpus. Industry-grade NLP corpus: ~25 KB per transcript average. Ideal for sentiment models, topic modeling, executive language change-detection, earnings-drift research keyed off transcript embedding similarity, and LLM fine-tuning on real corporate disclosure language.

Five years of sell-side analyst activity for the top 3,000 US-listed issuers, delivered as three complementary files: 85,777 individual upgrade/downgrade events, 73,147 individual price target changes, and 144,410 monthly aggregate strong-buy / buy / hold / sell / strong-sell counts. Each event row carries the analyst firm, previous and new grade or target, the stock price at the time of the call, and a link back to the originating news item. Read with pandas.readcsv(path, compression='gzip', parsedates=['publishedDate']). Useful for: analyst-momentum factors, upgrade/downgrade event studies, calibrating analyst accuracy by firm, building target-change feature stacks, and reconstructing consensus drift around earnings.

Environmental, Social, and Governance scores for the top 1,000 US-listed issuers by market cap (approximating the Russell 1000). Three complementary files: (1) esgdisclosures — 68,097 per-filing E/S/G/composite scores across 978 companies, linked back to the originating SEC form. (2) esgratings — 18,099 per-fiscal-year ESG risk-rating letters (A–F scale) plus industry rank, across 967 companies. (3) esgsectorbenchmark — 6,640 sector-level annual averages for benchmarking. ESG data of this quality is typically paywalled by MSCI / Sustainalytics; this dataset gives you an open, reproducible alternative. Read with pandas.readcsv(path, compression='gzip', parsedates=['date']). Useful for: ESG-tilted portfolio construction, sustainability research, regulatory disclosure analysis, and training models that need ESG features.

Five years of SEC Form 4 insider transaction filings for the top 918 US-listed companies by reporting history, covering 2021-01-04 through 2026-09-04. 601,965 transactions as a single gzipped CSV — one row per filed transaction. Columns include the filing date, transaction date, reporting insider name and role, transaction type (open-market buy/sell, option exercise, RSU vest, gift, etc.), shares transacted, price per share, total shares owned after the transaction, and a direct link to the EDGAR filing. Sourced from Financial Modeling Prep (Form 4 mirror of EDGAR). Read with pandas.readcsv(path, compression='gzip', parsedates=['filingDate', 'transactionDate']) or duckdb.read_csv('file.csv.gz'). Useful for: tracking insider sentiment, building cluster-buying signals, identifying executives unloading positions ahead of weakness, screen for management vs. board behavior divergence.

1,152,906 filer-concept-year facts for 16,041 SEC registrants across 15 US-GAAP concepts (revenue, cost of revenue, gross profit, operating income, net income, total assets, liabilities, shareholders' equity, cash, long-term debt, operating cash flow, capital expenditure, diluted EPS and shares outstanding) for calendar years 2009 to 2025, from the SEC's XBRL frames API as of 2026-09-06. Long format: one row per company, concept and calendar-year frame with CIK, entity name, state or country of incorporation, unit, period start and end, value and the accession number of the filing the fact came from, 250 frames fetched. Join to the SEC Company Tickers set on CIK.

Ten years of pre-computed quarterly financial ratios and metrics for the top 3,000 US-listed issuers, covering 103,013 ratio-period rows across 2,973 companies and 102,983 key-metric rows across 2,973 companies. Two complementary files in one listing: (1) keyratiosquarterly — profitability margins, turnover ratios, liquidity, solvency, leverage, valuation multiples (P/E, P/B, P/S, P/FCF, EV/EBITDA), per-share book values, dividend metrics, and tax/interest burdens. (2) keymetricsquarterly — market cap, enterprise value, EV multiples, return-on-capital family (ROA, ROE, ROIC, ROCE), working-capital cycles (DSO, DPO, DIO, cash conversion cycle), and capex/R&D/SBC intensity. Saves buyers weeks of feature engineering on top of raw IS/BS/CFS statements. Read with pandas.readcsv(path, compression='gzip', parsedates=['date']). Useful for: fundamental factor models, multi-factor backtests, screening by financial health, training ML models on pre-computed financial features.

Five years of ticker-tagged news articles for the top 1,500 US-listed issuers by market cap, covering 1,453,949 articles across 1,453 companies. Each row carries publication timestamp, publisher, site domain, headline, article body snippet, canonical URL, and image URL — all linked back to the underlying ticker symbol. Split into one CSV.GZ file per calendar quarter (21 files) for partial-history loading. Read with pandas.readcsv(path, compression='gzip', parsedates=['publishedDate']). Useful for: training news-sentiment models, event-study backtests around news catalysts, building ticker-mention frequency time series, fine-tuning LLMs on financial-news prose, and reconstructing the news flow around earnings, M&A, regulatory events, and macro shocks.