Lib Akshare

akshare

The widest free Chinese-market coverage by a large margin, and a pure scraper you cannot pin, cannot use for point-in-time anything, and cannot legally redistribute the data from.

Surface verified by ast-parsing __init__.py: 1,103 public interfaces. The scraper tells are in the dependency list — curl_cffi (TLS-fingerprint impersonation), mini-racer (embedded JS engine), bs4/lxml/html5lib. Upstreams: 东方财富 · 新浪 · 同花顺 · 金十 · 和讯 plus the exchange sites.

The trap that costs you money

You cannot pin a version, because akshare deletes its own PyPI history. 219 releases are listed but the oldest surviving release is 1.16.72 (2025-04-05). pip install akshare==1.12.x fails outright, so a requirements.txt written a year ago no longer resolves. Combined with ~2.3 releases/week, function signatures and returned column names move under you between installs.

There is no fix inside akshare. Vendor the data you need to Parquet. Do not plan to reproduce a result by pinning the library — plan to reproduce it from the bytes you saved.

Index constituents are current-only — every index backtest on them is biased

ak.index_stock_cons_csindex(symbol=...) has no date parameter. It returns today's HS300/CSI500 membership and nothing else. Backtesting "the HS300" with it means holding, throughout history, exactly the names that were promoted into the index later — textbook inclusion bias, and akshare cannot tell you it happened.

Use tushare index_weight / index_member, baostock's dated (HS300/SSE50/CSI500-only) lists, or a licensed vendor.

Adjustment: the default is safe — keep it that way

stock_zh_a_hist(..., adjust="") returns raw (不复权) prices, verified in source. That is the correct base for limit-up, tick-size and lot-size logic.

• adjust="hfq" — anchored at listing, append-only, reproducible. Use this for signals.
• adjust="qfq" — anchor is now, so re-running the same query next month returns different history. Never persist it.
• stock_zh_a_daily(..., adjust="hfq-factor") — the Sina path uniquely returns the adjustment factors themselves, which is the right way to keep raw prices and roll your own convention. Its own docstring warns 大量抓取容易封 IP.

No point-in-time fundamentals, and no unified schema

Financial statements come back keyed on 报告期 with no usable 公告日. Joining on 报告期 leaks up to four months of future information (the annual-report deadline is 30 April).

There is also no common return type across the 1,103 functions — column names, dtypes and row ordering differ per upstream page and change when that page changes. Normalize at the boundary and never index positionally. Suspension and limits are indirect: stock_zh_a_stop_em for 停牌, and the limit pools (stock_zt_pool_em plus the 跌停/炸板/强势 variants) substitute for real high_limit/low_limit fields. Delisted price history is unverified — test it for your universe before claiming survivorship-free coverage.

MIT covers the scraper, not the data

akshare scrapes 东方财富 / 新浪 / 同花顺 / 腾讯, whose ToS prohibit automated bulk extraction and commercial redistribution. The MIT licence grants nothing regarding the scraped data. akshare's own docs describe the data as for academic research purposes and warn of commercial risk. 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 citing data-security policy.

Free scrapers for research, prototyping and personal use. Anything commercial, client-facing or redistributed needs Wind, Choice, RiceQuant, JQData or tushare's paid tier. And rate-limit from the first line of code: curl_cffi is in the tree precisely to impersonate a browser, and akshare's own source warns you will get your IP banned.

Minimal correct call

import akshare as ak

# raw is the default and the correct base for limit-up / tick-size / lot-size logic
raw = ak.stock_zh_a_hist(symbol="000001", period="daily",
                         start_date="20200101", end_date="20260901", adjust="")

# hfq for signals: anchored at listing, append-only, reproducible
hfq = ak.stock_zh_a_hist(symbol="000001", period="daily",
                         start_date="20200101", end_date="20260901", adjust="hfq")

# best of both: raw prices + the factors, so you own the convention (Sina path only)
fct = ak.stock_zh_a_daily(symbol="sz000001", adjust="hfq-factor")

# never do this for anything you store or backtest on:
# ak.stock_zh_a_hist(..., adjust="qfq")   # anchor = today; history is rewritten every dividend

# and never build a historical universe from this: it is TODAY's membership, no date parameter
# ak.index_stock_cons_csindex(symbol="000300")

See also

• ../../../fin-china/skills/china-ashare-data/SKILL.md — 复权, 涨跌停, 停牌, 公告日 vs 报告期
• ../../../fin-china/skills/china-ashare-data/references/akshare.md — the verified reference card
• ../../../fin-china/skills/china-ashare-data/references/_source-matrix.md — capability matrix
• ../lib-tushare/SKILL.md — point-in-time fundamentals and dated index membership

Where this sits

This file is the deep dive on one library and assumes the choice is already made. For which library to pick, how it compares with the alternatives, and the traps that span several of them, the entry point is the domain skill china-ashare-data (../../../fin-china/skills/china-ashare-data/SKILL.md).