China Ashare Data

China A-share data

The Chinese ecosystem's defaults are more dangerous than the Western one's: three of the most popular libraries default to forward-adjusted (前复权) prices, which are retroactively rewritten and therefore look-ahead contaminated. Read §2 before anything else.

Most facts below are ✅ verified by downloading the wheel and reading the source, not from docs.

1. Pick a source

Do not use: pytdx (archived 2020, no licence), Ashare (no licence — you cannot legally ship it), adata for backtests (§2), qstock for anything you build on (18 months stale).

2. 🚨 复权 — the qfq look-ahead bias

Three conventions:

• 不复权 (raw) — the actually traded price. Correct for limit-up/down logic, tick sizes, lot sizing.
• 后复权 hfq — anchored at the listing date: adj = raw × factor. A new dividend only appends a factor; history never changes. Append-only, stable, reproducible.
• 前复权 qfq — anchored at now: adj = raw × factor / factor(anchor).

Why qfq is poison: the anchor moves. Every new corporate action rescales the entire history. So (a) the same query run a month later returns different historical prices — non-reproducible; and (b) a qfq price at t depends on events after t — it could not have been computed at t.

Defaults, all ✅ verified in source:

🚨 tushare's qfq is worse than the standard kind. Verified in tushare/pro/data_pro.py:

if adj == 'hfq': data[col] = data[col] * data['adj_factor']
if adj == 'qfq': data[col] = data[col] * data['adj_factor'] / float(fcts['adj_factor'][0])

fcts is returned newest-first, so [0] is the factor on the last date of your query window — the same bar has different values depending on what end_date you asked for. Two further subtleties in that function: adj_factor is bfill()-ed, and pre_close = close.shift(-1) is only correct because rows arrive descending — sort ascending first and pre_close, change and pct_chg all silently break.

Rule: hfq, or raw + factors. Never qfq. Render qfq at display time only; never persist it.

3. 涨跌停, ST, T+1 — fills that could not have happened

A bar closing at limit-up generally could not have been bought. Filling at close on limit-up days is the second-largest source of fake alpha in A-share research, after survivorship.

ST is a name change, not a code change — a ticker's limit rule changes mid-history with no identifier change. Logic that assumes ±10% mislabels every ST bar.

Only jqdatasdk (high_limit/low_limit/paused fields, get_preopen_infos) and the commercial vendors give exact bounds. Free substitute: akshare's limit-up pools (stock_zt_pool_em plus 跌停/炸板/强势 variants). Deriving limits from prev_close × (1±pct) is error-prone because of the board matrix, ST's ±5%, and tick rounding.

T+1 settlement: shares bought today cannot be sold until the next session (cash is T+0). Any intraday-reversal strategy on A-shares is unimplementable. Futures are T+0 — a mixed equity/futures backtest needs two settlement models.

北交所 is the weakest coverage link. rqdatac needs enable_bjse=True — default False, so BSE is silently absent. Verify BSE explicitly before claiming a "full A-share" universe.

4. 停牌 (suspension) → zero-volume bars

Chinese suspensions run for months. Sources handle them three incompatible ways: omit the bar / emit volume=0 with a forward-filled price / emit NaN. Forward-filled bars create spurious zero returns that deflate measured volatility; omitted bars misalign multi-asset panels.

Explicit control: jqdatasdk (skip_paused, fill_paused, paused), baostock (tradestatus per bar + query_suspended_stocks()), akshare (stock_zh_a_stop_em). The scrapers give you nothing — inferring from volume == 0 also catches genuinely illiquid bars.

5. 公告日 vs 报告期 — the fundamentals look-ahead

Every filing has 报告期 / end_date (the period described) and 公告日 / ann_date / info_date (when it was published). Annual reports appear up to four months after period end (deadline 30 April). Joining on 报告期 leaks up to four months of future information — alone enough to turn a worthless value factor into a spectacular backtest.

Filings are also aggressively restated: 业绩预告 → 业绩快报 → 正式财报 → later restatements.

🚨 rqdatac's PIT call has a trap of its own: default statements='latest', date=None returns the most recently restated record. Its own docstring example returns a 2019 restatement for a 2018 quarter. You must pass date=<as-of>.

🚨 tushare's stock_basic defaults to list_status='L' — omit it and your universe is silently survivorship-biased. Delisting in China went from rare to common after the 2020 退市新规, so the bias is time-varying, which is worse than a constant one.

6. Sessions and timezone

Asia/Shanghai (UTC+8), no DST. Equity session 09:30–11:30, 13:00–15:00, plus 09:15–09:25 集合竞价 and 14:57–15:00 closing auction. Exactly 240 minute bars in a normal session.

🚨 The 90-minute lunch break breaks naive resampling. pandas.resample('1min') or a wall-clock rolling window silently spans 90 minutes of non-existent time. Compute rolling windows over bar index, never wall-clock.

Futures 夜盘 runs ~21:00 to 23:00/01:00/02:30 → a futures trading day starts the previous calendar evening. A classic off-by-one. tqsdk and rqdatac handle it; scrapers do not.

7. Licence and legality — the data, not the code

All code licences here are permissive (MIT / BSD-3 / Apache-2.0). Two exceptions that are stricter than GPL: pytdx and Ashare have NO licence file = all rights reserved under Berne. You may not legally redistribute or ship them.

🚨 The real risk is the data. akshare, efinance, adata, qstock, easyquotation and Ashare scrape 东方财富 / 新浪 / 同花顺 / 腾讯, whose ToS prohibit automated bulk extraction and commercial redistribution. The MIT licence on the scraper grants nothing regarding the scraped data. akshare's own docs say the data is "only for academic research purposes" and warn of "commercial risks". 反不正当竞争法 Art. 12 and 数据安全法 have both been applied to systematic financial scraping — and Microsoft's own qlib pulled its official China dataset "due to more restrict data security policy", a live example of the risk materializing.

Rule: free scrapers for research, prototyping and personal use. Anything commercial, client-facing or redistributed needs a licensed vendor (Wind, Choice, RiceQuant, JQData, or tushare's paid tier).

Operational: these get IP-banned. akshare's own source warns 大量抓取容易封 IP; adata ships a proxy hook; akshare depends on curl_cffi for TLS-fingerprint impersonation. Plan rate limiting and caching from the start.

8. Two more things that will bite

• 🚨 akshare purges its own PyPI history. 219 releases are listed but the oldest surviving is 1.16.72 (2025-04-05) — you cannot pip install akshare==1.12.x. Combined with ~2.3 releases/week (the cadence is the maintenance model, since ~85% of changes are scraper fixes), vendor the data you need rather than pinning a version.
• 🚨 tushare sends your token over plaintext HTTP. ✅ verified in 1.4.29 source: __http_url = 'http://api.waditu.com/dataapi'. No HTTPS option is shipped.
• ⚠️ tushare's GitHub repo last pushed 2024-03-13, but PyPI shipped releases in March 2026 — the public repo is not the source of truth for the package.
• 🚨 baostock's new 0.9.x functions are not re-exported in __init__.py. Import the submodule: from baostock.security.sectorinfo import query_terminated_stocks.

9. Canonical safe snippets

# akshare — safe default (raw), or explicit hfq
import akshare as ak
raw = ak.stock_zh_a_hist("000001", "daily", "20200101", "20260901", adjust="")
hfq = ak.stock_zh_a_hist("000001", "daily", "20200101", "20260901", adjust="hfq")

# tushare — universe INCLUDING delisted, and PIT fundamentals
import tushare as ts
pro = ts.pro_api("TOKEN")
live = pro.stock_basic(list_status="L", fields="ts_code,name,list_date,delist_date")
dead = pro.stock_basic(list_status="D", fields="ts_code,name,list_date,delist_date")  # required!
bars = ts.pro_bar(ts_code="000001.SZ", adj="hfq", start_date="20200101")  # never adj="qfq"
inc  = pro.income(ts_code="000001.SZ", fields="ts_code,ann_date,f_ann_date,end_date,update_flag,revenue")
# filter f_ann_date <= as_of for a genuine point-in-time view

# baostock — free delisted list + raw/hfq bars
import baostock as bs
from baostock.security.sectorinfo import query_terminated_stocks
bs.login()
rs = bs.query_history_k_data_plus(
    "sh.600000", "date,code,open,high,low,close,volume,tradestatus,isST",
    start_date="2020-01-01", end_date="2026-09-01", frequency="d", adjustflag="1")  # 1 = hfq
delisted = query_terminated_stocks()
bs.logout()

10. Reference files

references/ holds one file per library with exact versions, source-verified signatures, traps and snippets. Grep before guessing:

grep -ril "复权\|adjust" plugins/fin-china/skills/china-ashare-data/references/
grep -i -A6 "TRAP" plugins/fin-china/skills/china-ashare-data/references/tushare.md

❓ Not verified — state as unknown, do not guess

tushare 积分 per-minute rate table and RMB prices (thresholds ✅, frequencies ❓) · commercial pricing for Wind / JQData / rqdatac / TqSdk Pro / Choice (none publish prices; every figure in circulation is secondhand) · JoinQuant's post-2021 regulatory status (site geo-blocked) · whether akshare returns price history for delisted tickers · what baostock's new API key unlocks · Futu / Tiger HK SDKs.

Per-library deep dives

The optional fin-libraries plugin carries a dedicated skill for each library below. Load one only after this skill has told you which library you want:

• lib-akshare — akshare
• lib-tushare — tushare