Free, research-grade, high-frequency intraday data — 1-minute OHLCV bars for 1,391 U.S. stocks and ETFs. Documented, version-controlled, and updated daily.
One-minute OHLCV bars for 1,391 U.S. stocks and ETFs (equities and exchange-traded funds), from December 2002 through the present.
Two cleaning versions (Raw and Clean) so you can choose the level of processing appropriate for your research. Twenty-five pre-computed academic variables per ticker per day. Full methodology documentation.
A free, citable alternative to TAQ and CRSP for intraday research — minute-by-minute history for backtesting, volatility estimation, and market-microstructure studies.
Updated daily. No subscription. No paywall. Licensed under CC BY 4.0.
# Python — load any ticker in seconds
import pandas as pd
df = pd.read_parquet("AAPL_clean.parquet")
print(df.head())
# datetime Open High Low Close Volume
# 2002-12-30 09:30:00 0.98 0.99 0.98 0.98 842900
# 2002-12-30 09:31:00 0.98 0.99 0.98 0.99 521400
# ...Data as received from the source. No outlier removal, no gap-filling. Prices are split/dividend adjusted. 1.5+B bars.
Best for: Market microstructure research, missingness analysis, studying the data itself.
Nine-step cleaning pipeline applied: outside-hours removal, non-positive prices, OHLC violations, duplicate bars, Brownlees-Gallo outlier filter. Gaps preserved. 1.5+B bars.
Best for: Volatility estimation, spread measurement, jump detection — most empirical finance.
Note: A gap-filled version is not distributed. Gap-filling (LOCF) introduces biases documented in the methodology — researchers who need a regular grid can apply LOCF to the Clean version themselves.
Computed daily for each ticker in each cleaning version. Ready to use in your research.
Realized variance (1-min, 5-min), bipower variation, Parkinson range, Yang-Zhang OHLC
Roll (1984) implied spread, Corwin-Schultz (2012) high-low spread
First-order return AC(1), variance ratio VR(5), VR(10)
BNS z-statistic, jump indicators at 1% and 5% significance
Amihud illiquidity, daily dollar volume, share volume, observed trade count
Gap rate, observed/filled bar counts, longest gap, bars since last trade
Download individual tickers or pre-packaged bundles (S&P 500, Nasdaq 100, by sector) as Parquet or CSV. Click and go — no account needed for basic downloads.
Browse DownloadsProgrammatic access to any ticker, date range, and version. JSON, CSV, or parquet. Free API key with 100 downloads/minute. Python, R, and Stata examples provided.
API DocsConnect Claude, Cursor, Gemini, or ChatGPT and ask for data in plain English — the MCP server finds the series, downloads it, and cites it. No code required, and — unlike the paid market-data APIs — completely free.
MCP GuideFull dataset dump — all 1,391 tickers, all versions, all timeframes. Parquet format. Updated daily.
Full Dataset| Feature | HF Data Library | CRSP/TAQ | Yahoo Finance | Polygon.io |
|---|---|---|---|---|
| Price | Free | $25,000+/yr | Free | $199+/mo |
| Frequency | 1-minute bars | Tick-level | Daily only | 1-minute bars |
| Cleaning versions | 2 versions | 1 version | None | None |
| Cleaning documentation | Full pipeline | Minimal | None | None |
| Academic variables | 25 measures | None | None | None |
| Data quality scores | Per-ticker | No | No | No |
| REST API | Free | No | Unofficial | Paid |
| AI access (MCP) | Free | No | Unofficial | Paid |
| DOI / Citable | Zenodo DOI | No | No | No |
| License | CC BY 4.0 | Restrictive | ToS restricted | Commercial |
| Updated | Daily | Quarterly | Daily | Real-time |
Among these, the HF Data Library is the only source with its own free MCP server for AI assistants — Polygon.io's requires a paid plan, and the others don't offer one. See the MCP guide →
Read more: a free TAQ & CRSP alternative for intraday research →
Yes. The entire dataset is free under the Creative Commons CC BY 4.0 license — no subscription and no paywall. You only need to attribute the HF Data Library.
1-minute OHLCV bars for 1,391 U.S. stocks and ETFs from December 2002 to the present (about 1.5 billion bars), in two cleaning versions (Raw and Clean), plus 25 pre-computed academic variables per ticker per day.
1,391 U.S. stocks and ETFs, including large caps and major index members such as AAPL, MSFT, TSLA, SPY, and QQQ. See the Tickers page for the full list.
Apache Parquet and CSV files for download, and a free REST API that returns JSON, CSV, or Parquet.
December 30, 2002 to the present for most tickers. Coverage varies by ticker — newer listings start later, and a handful of series begin earlier.
Yes. A free REST API is available with a free key and a 100 downloads-per-minute limit, returning data in JSON, CSV, or Parquet.
Yes. Connect the free ElkassabgiData MCP server to Claude, Cursor, Gemini, or ChatGPT and ask for data in plain English — it finds the series, downloads it, and cites it. The MCP guide has step-by-step setup for each client.
It is a free, citable option for intraday U.S. equity research without WRDS — 1-minute bars plus 25 pre-computed measures. It is not a tick-level replacement for TAQ, since the bars are aggregated to one minute. See the full comparison →
Cite the dataset by its Zenodo DOI, 10.5281/zenodo.19501605. A formatted citation, BibTeX, and a CITATION.cff file are on the Cite page.