← All tools · Guides

AKShare: China A-share market data into SQLite, updated every day

Hands-on, part 2. Download, clean, and append China A-share bars so a backtest has a fresh table every weekday.

How to choose: To store China A-share daily bars on your own computer, use AKShare and append them to SQLite. After pip install, call stock_zh_a_hist with adjust="qfq" for one symbol, rename the Chinese columns to lowercase English, and each run fetch only the dates after MAX(date). The full-market snapshot is stock_zh_a_spot_em. The index series is stock_zh_index_daily. Run cron at 18:30 on weekdays, after the session, and sleep 0.5 seconds between symbols. Fees and the free tier for Tushare versus AKShare are a different article. A US-stock pipeline with yfinance or EODHD is a different article. This page only walks AKShare market data into a local database.

Who it is for: Someone who writes a little Python and wants China A-share daily bars, a full-market snapshot, and an index series in a local SQLite file, updated every weekday, with no registration key. The path is the three calls, the column rename, the incremental write, and a job after 18:30.

Who it is not for: Someone who needs a commercial data licence, a data contract that will not change, or US prices. AKShare aggregates public sites such as Eastmoney and Sina. An interface can change. The stored history is for personal research. This page is not the tool-order guide, not the site FAQ, and not the Tushare fee comparison. If you do not yet have a machine that stays on, start with the cloud quant desk, the same article in English and Chinese.

The same walkthrough in Chinese is AKShare: market data into a database. It is part 2 of that series. The previous part, a machine that stays on, is the cloud quant desk. The same article in Chinese is the Chinese cloud-setup guide. Fees, the free tier, and why an interface can fail are how to choose between Tushare and AKShare. That comparison is not this article. Which of yfinance, AKShare, OpenBB, and EODHD to use is how to choose a data source. That guide is not this article. Giving those bars to a research agent is market data and an LLM API. That note is not this article. A US-stock pipeline that was actually run is the daily price pipeline. The next part of this series, a first strategy on these bars, is the first backtrader strategy.

Facts below follow the Chinese walkthrough. The commands were not re-run for this English page. This site has no commercial relationship with AKShare. This is not investment advice. The interfaces are public aggregations. Cross-check anything you will rely on.

01 · What AKShare is, and who it fits

AKShare is an open-source Python library of finance interfaces. It aggregates public sources such as Eastmoney and Sina. It is free and needs no registration key. For a person doing their own research, it is the short path from an idea to a table. The cost is that those sources are not an official contract. An interface can change. The last section is the list of ways that shows up.

pip install akshare pandas -i https://pypi.tuna.tsinghua.edu.cn/simple

The -i flag is the Tsinghua PyPI mirror from the Chinese page. pip install akshare pandas is the same install without that mirror.

02 · The three calls you will actually use

import akshare as ak

# 1) Daily bars for one symbol (forward-adjusted, the backtest convention). Eastmoney source.
df = ak.stock_zh_a_hist(symbol="000001", period="daily",
                        start_date="20230101", end_date="20250801", adjust="qfq")

# 2) Full-market snapshot (last price, change, market cap). Works during the session and after.
spot = ak.stock_zh_a_spot_em()

# 3) Index daily bars
idx = ak.stock_zh_index_daily(symbol="sh000001")

print(df.head())   # the columns come back with Chinese names

Chinese column names are awkward downstream. Rename them to lowercase English. That is step one of every AKShare pipeline. The strings below are the Chinese names the frame returns:

import pandas as pd

RENAME = {"日期": "date", "开盘": "open", "最高": "high",
          "最低": "low", "收盘": "close", "成交量": "volume"}

df["date"] = pd.to_datetime(df["日期"])
df = df.rename(columns=RENAME).set_index("date")[["open", "high", "low", "close", "volume"]]

03 · Incremental writes: SQLite is the personal default

One file, nothing to operate, and pandas reads and writes it in two lines. The Chinese page says that is enough into the tens of millions of rows. The rule is to fetch only the stretch the database does not have yet:

import sqlite3, time
import akshare as ak

DB = "a_daily.db"

def update_one(code: str, conn):
    start = "20230101"
    # If the code is already stored, start the day after the last date.
    exist = conn.execute(
        "SELECT MAX(date) FROM daily WHERE code=?", (code,)).fetchone()[0]
    if exist:
        start = (exist + pd.Timedelta(days=1)).strftime("%Y%m%d")

    df = ak.stock_zh_a_hist(symbol=code, period="daily",
                            start_date=start, end_date="20991231", adjust="qfq")
    if df.empty:
        return
    df["date"] = pd.to_datetime(df["日期"])
    df = df.rename(columns={"开盘": "open", "最高": "high", "最低": "low",
                            "收盘": "close", "成交量": "volume"})
    out = df[["date", "open", "high", "low", "close", "volume"]]
    out["code"] = code
    out.to_sql("daily", conn, if_exists="append", index=False)

conn = sqlite3.connect(DB)
conn.execute("""CREATE TABLE IF NOT EXISTS daily
    (code TEXT, date TIMESTAMP, open REAL, high REAL,
     low REAL, close REAL, volume REAL)""")
update_one("000001", conn)   # in a real project, loop the universe
conn.close()

pd is the pandas import from the rename step. Two details the loop still needs: time.sleep(0.5) between symbols, so the source is not hammered, and the universe itself from ak.stock_info_a_code_name().

04 · Update it every weekday

Save the script as update_daily.py and hang it on cron. The machine has to stay on:

crontab -e
# Weekdays at 18:30, after the session, when the daily bar has settled
30 18 * * 1-5 cd /home/quant/jobs && python update_daily.py >> logs/data_$(date +\%F).log 2>&1

05 · Pitfalls, each one already paid for

• A bar pulled during the session is unfinished. The daily interface, called before the close, returns an incomplete candle for today. Put the job after 18:30.
• Rate limits and bans. Sleep inside the loop. On a failure, back off and retry. An immediate retry makes the ban worse.
• Adjustment. Backtests use qfq, forward-adjusted prices. Those prices drift across the whole history when a dividend or a split lands. Refresh the stored history every few months, or the old prices quietly go stale.
• Interfaces change. The project is active. An upgrade can rename a function or switch a source. Pin the version in requirements.txt, and rerun a check before you upgrade.
• Delisted names go missing. Some calls return only names that are still listed. A full-market backtest then has survivor bias. Say so when you quote the result.