vessel888/china-etf-1min-ohlcv

China Exchange-Traded Funds 1-Minute OHLCV

Minute-level OHLCV bars for selected exchange-listed Chinese ETFs. The release uses a stable Parquet schema, one canonical file per instrument, and machine-readable coverage reports.

Dataset summary

This snapshot contains 4,058,681 rows for 7 instruments across selected exchange-listed Chinese ETFs. It covers 2015-01-05 09:30:00 through 2026-08-20 15:00:00. Prices are unadjusted. Volume is stored in shares, turnover in CNY, and timezone-naive timestamps are interpreted as Asia/Singapore (UTC+8).

Item	Value
Schema version	1
Instruments with bars	7
Minute-bar rows	4,058,681
Canonical Parquet files	7
Normalized data size	0.05 GB
First timestamp	2015-01-05 09:30:00
Last timestamp	2026-08-20 15:00:00
Snapshot build time	2026-08-21T16:41:46+08:00
Calendar/listing instrument-days before market-status audit	16,842
Observed instrument-trading days	16,841
Instrument-days without observed bars	1
Raw calendar/listing coverage before market-status audit	99.9941%
Instruments with no-bar days before market-status audit	1
Market-wide missing trading days inside the published span	0

Scope: this repository contains only the selected ETFs represented in the published inventory; it is not a complete universe of China-listed ETFs. A-share equities are excluded and published separately.

Dataset Viewer and subsets

The canonical data uses one Parquet file per instrument. Asking the Hub to inspect all 7 remote files as one Viewer split can time out. The explicit subsets below keep the Viewer usable without changing or duplicating the canonical dataset.

Subset	Contents	Rows
bars_sample (default)	First up to 64 chronological bars from every published instrument; deterministic and not synthetic	448
instrument_coverage	One row per published instrument with path, time range, record count, and trading-day coverage	7
missing_intervals	Consecutive eligible trading-day ranges with no observed bars	1
market_calendar	Every calendar date labeled as trading day, weekend, or exchange closure, with market-wide bar availability	4,250
absent_instruments	Eligible universe members with no published canonical file	0

The Viewer sample spans 7 instruments and is 0.01 MB. It exists only for browser inspection. The complete 4,058,681-row dataset remains under data/etf_1m/.

load_dataset("neigezhu/china-etf-1min-ohlcv") loads the default bars_sample subset. Use the file-level or full-snapshot methods below for the canonical bars.

Load a Viewer subset

from datasets import load_dataset

sample = load_dataset("neigezhu/china-etf-1min-ohlcv", "bars_sample", split="train")
coverage = load_dataset("neigezhu/china-etf-1min-ohlcv", "instrument_coverage", split="train")

Load one canonical instrument

from huggingface_hub import hf_hub_download
import pyarrow.parquet as pq

path = hf_hub_download(
    repo_id="neigezhu/china-etf-1min-ohlcv",
    filename="data/etf_1m/SH/510300.parquet",
    repo_type="dataset",
)
bars = pq.read_table(path)

Download and scan the complete snapshot

from pathlib import Path
from huggingface_hub import snapshot_download
import pyarrow.dataset as ds

root = Path(snapshot_download(
    repo_id="neigezhu/china-etf-1min-ohlcv",
    repo_type="dataset",
    allow_patterns=["data/etf_1m/*/*.parquet", "metadata/*"],
))
bars = ds.dataset(root / "data" / "etf_1m", format="parquet")

For reproducible research, pass a Hub commit SHA as revision= to hf_hub_download or snapshot_download.

Repository layout

data/etf_1m/{exchange}/{symbol}.parquet
viewer/bars_sample.parquet
viewer/instrument_coverage.parquet
viewer/missing_intervals.parquet
viewer/market_calendar.parquet
viewer/absent_instruments.parquet
metadata/coverage_by_instrument.csv
metadata/missing_intervals.csv
metadata/market_calendar.csv
metadata/absent_instruments.csv
metadata/summary.json
LICENSE
README.md

Each canonical instrument file is sorted by timestamp. Its logical key is (exchange, symbol, timestamp). Symbols remain strings so leading zeroes are preserved.

Bar schema and semantics

Field	Parquet type	Unit / meaning
symbol	string	Security identifier without exchange suffix
exchange	string	SH, SZ, or BJ
timestamp	timestamp[us]	Timezone-naive minute label interpreted as Asia/Singapore (UTC+8)
open	float64	Unadjusted opening price, CNY per share
high	float64	Unadjusted high price, CNY per share
low	float64	Unadjusted low price, CNY per share
close	float64	Unadjusted closing price, CNY per share
volume	int64	Traded shares during the minute
turnover	float64	Traded value during the minute, CNY

• Frequency: 1 minute.
• Regular session labels: 09:30–11:30 and 13:00–15:00.
• Price adjustment: none; corporate-action adjustment factors are not included.
• Storage: Zstandard-compressed Parquet with statistics and bounded row groups.

Coverage and quality reports

metadata/coverage_by_instrument.csv is the canonical inventory. It records each file path, observed range, row count, expected and observed trading days, and coverage status. metadata/missing_intervals.csv groups eligible trading-day ranges with no observed bars. metadata/market_calendar.csv labels every date as trading_day, weekend, or exchange_closed, and records whether any published instrument has bars on that date. metadata/absent_instruments.csv lists eligible instruments without a published file. metadata/summary.json provides snapshot-level counts for automated checks.

Coverage is evaluated against an exchange trading calendar and each instrument's listing range. Weekends and exchange-declared closures, including statutory holiday closures, are excluded from expected trading days. Dates before listing and after known delisting are also excluded.

The value 1 instrument-day without observed bars is an aggregate over instruments, not a count of distinct calendar dates. If 100 instruments have no bars on one open day, that contributes 100 no-bar instrument-days. Within the published span, 0 exchange trading days have no bars for any instrument.

This snapshot does not publish an instrument-day suspension audit, so a no-bar day must not be interpreted as a verified collection gap from this repository alone.

Construction and validation

This snapshot is built by normalizing collected minute-bar files to the public schema, then sorting and deduplicating records by instrument and timestamp. Coverage reports are regenerated from the published Parquet files. This ETF release does not claim independent multi-source corroboration.

Limitations

• This is a historical snapshot, not a real-time feed.
• Coverage varies by instrument and is not gap-free; inspect instrument_coverage before research or backtesting.
• Prices are unadjusted and the release does not include corporate actions or adjustment factors.
• Minute OHLCV does not contain orders, individual trades, order-book state, participant identities, or latency information.
• Provider-level provenance is intentionally not exposed, so it cannot be independently audited from the repository alone.
• Floating-point prices and turnover should not be treated as exact decimal accounting values.

Related dataset

For A-share equities, use the separate China A-Share Equities 1-Minute OHLCV dataset. The schemas are compatible, but repository paths, universes, row counts, and coverage metrics are independent.

Versioning, license, and citation

• Dataset schema version: 1.
• License: Apache License 2.0.
• Updates preserve the canonical path and field contract. Consumers that require an immutable snapshot should pin a Hub commit SHA.

@dataset{neigezhu_china_etf_1m_ohlcv,
  author    = {neigezhu},
  title     = {China Exchange-Traded Funds 1-Minute OHLCV},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/neigezhu/china-etf-1min-ohlcv}
}

---

中文说明

数据集概览

本快照包含选定的中国上市 ETF，共 7 个证券、4,058,681 条 1 分钟 OHLCV，时间范围为 2015-01-05 09:30:00 至 2026-08-20 15:00:00。价格不复权，成交量单位为股，成交额单位为人民币。Parquet 的 timestamp 不携带时区，统一按 Asia/Singapore（UTC+8）解释。

范围说明：本仓库只包含发布清单中的选定 ETF，并非中国上市 ETF 全量库；A 股股票不在本仓库内，另行发布。

Dataset Viewer

主数据保持“每个证券一个 Parquet 文件”，共有 7 个文件。Hugging Face 若把这些远程文件作为一个 Viewer split 逐个读取元数据会超时，因此仓库显式配置了五个可用子集：

• bars_sample：默认子集；每个已发布证券取最早不超过 64 条记录，共 448 条。样本是确定性的真实记录，不是合成数据。
• instrument_coverage：每个已发布证券一行，包含文件路径、记录数、起止时间和交易日覆盖。
• missing_intervals：连续缺失交易日区间。
• market_calendar：逐日标记交易日、周末或交易所休市日，并标记当天全市场是否至少有一条分钟数据。
• absent_instruments：应在证券范围内但没有主数据文件的证券。

Viewer 只用于浏览。完整的 4,058,681 条记录仍位于 data/etf_1m/。直接调用 load_dataset("neigezhu/china-etf-1min-ohlcv") 会读取默认的 bars_sample，完整数据应使用上面的单证券下载或全快照下载方式。

数据约定

• 逻辑主键：exchange + symbol + timestamp。
• 单个证券文件按 timestamp 升序排列。
• 交易所代码：SH、SZ、BJ。
• 常规交易时段：09:30–11:30、13:00–15:00。
• 价格：人民币/股，不复权。
• 成交量：股。
• 成交额：人民币。
• 时间：无时区 Parquet 时间戳，按 Asia/Singapore（UTC+8）解释。

构建与质量

本快照将已采集的分钟文件规范化到公开字段，并按证券和时间戳排序、去重；覆盖报告由最终发布的 Parquet 文件重新生成。本 ETF 版本不声明经过独立多源交叉核验。

覆盖率按交易所交易日历和证券上市区间计算。周末、法定节假日休市和交易所公告休市日不进入应有交易日；上市前和已知退市后的日期也不计入。

当前共有 16,842 个日历/上市区间内证券交易日、16,841 个已观察证券交易日、1 个无分钟记录证券日。这个无记录数是“证券 × 交易日”的合计，不是不同自然日数量：同一个交易日有 100 只证券无记录会计 100。发布区间内，全市场所有证券同时没有数据的交易日为 0 天。

研究和回测前应同时检查 instrument_coverage、missing_intervals 和 market_calendar。本快照未发布逐证券日停牌核验表，因此不能仅凭无分钟记录判定为采集缺口。

局限

• 这是历史快照，不是实时行情。
• 不同证券的覆盖范围不同。
• 数据不复权，不包含公司行动和复权因子。
• 分钟 OHLCV 不包含逐笔成交、订单簿、参与者身份或延迟信息。
• 仓库不披露数据提供方名称，因此不能仅从公开文件独立审计提供方级别的来源。

相关数据集

A 股股票数据见独立的 China A-Share Equities 1-Minute OHLCV。两套数据字段兼容，但仓库路径、证券范围、记录数和覆盖指标彼此独立。

版本与许可

• schema 版本：1。
• 许可：Apache License 2.0。
• 需要严格复现时，请在下载接口中固定 Hugging Face commit SHA。