A-share top-trader boards with Python: four approaches

Option 1: call the HTTP API directly (recommended)

This is a paid endpoint (Pro tier and up): one request with your key returns the shared envelope `{ ok, endpoint, tier, elapsed_ms, source, data }`. The `type` parameter picks the board: `institution` / `hotmoney` / `activeseat`.

Inside the response, `data` is a human-readable Markdown table while `structured` is the same data already parsed into an array of objects — use `structured` in code instead of parsing Markdown.

import requests

KEY = "ct-your-key"                     # the top-trader endpoint needs a key (Pro tier and up)
r = requests.get(
    "https://api.ashareapi.com/v1/lhb",
    headers={"Authorization": "Bearer " + KEY},
    params={"type": "institution"},     # institution / hotmoney / activeseat
    timeout=30,
)
body = r.json()
if not body.get("ok"):
    raise RuntimeError(body)            # upstream failure returns ok=false and is NOT counted

for row in body["structured"][:3]:
    net = float(row["netBuyAmt"])       # values are strings, convert before doing math
    print(row["name"], row["instBuyBranchCount"], "institutions",
          "net buy", round(net / 1e8, 2), "x1e8 CNY")

• `structured[].code` carries the market prefix (`sz301234`) · `tdDays` = days on the board · `instBuyBranchCount` = number of buying institutions
• `instBuyAmt` / `netBuyAmt` are in CNY (not 10k CNY) · `instBuyRate` / `netBuyRate` are percentages (`20` means 20%)
• The response also carries `tables` (the Markdown tables parsed into 2D arrays), identical in content to `structured`

What the three boards are

One live sample (2026-09-23): the institution board returned 34 rows and active seats 161. Top institution entry: Wuzhou Medical (4 institutions bought 135M CNY, net 33M). Top active seat: the Shanghai Stock Connect seat, 345M CNY bought.

In active seats, `code` and `stockName` are both semicolon-separated lists with matching order — e.g. `sh600127;sh600664;sh688105` maps to `Jinjian Cereals;Harbin Pharma;Novogene`. Just `split(";")`.

Option 2: Tushare (top_list / top_inst)

Tushare ships official board endpoints with stable data, but you must sign up, get a token and hold enough credits: daily detail `top_list` needs 2000+ credits, institutional detail `top_inst` needs 5000+ (per Tushare official docs).

It returns flat detail rows (which seat bought how much of which stock) — there is no pre-grouped institution/hot-money board, so you aggregate yourself.

# pip install tushare
import tushare as ts

pro = ts.pro_api("your-token")
df = pro.top_list(trade_date="20260923")     # daily detail (needs 2000+ credits)
print(df[["ts_code", "name", "net_amount", "reason"]].head())

• When not to use it: you just want "what did institutions buy today" — flat detail still needs aggregation, a board endpoint is faster
• When it fits: you already run historical research on Tushare and need boards joined with financials and quotes

Option 3: AkShare (stock_lhb_detail_em)

AkShare needs no token and scrapes public East Money pages; the board entry is `stock_lhb_detail_em(start_date, end_date)`, with extra Sina-backed interfaces for daily detail, institutional seats and branch statistics.

The cost is that upstream redesigns break it and you maintain the upgrades (most of its issue tracker). Fine for one-off research; a long-running job needs someone watching.

# pip install akshare
import akshare as ak

df = ak.stock_lhb_detail_em(start_date="20260901", end_date="20260923")
print(df.head())

Option 4: scrape East Money / exchange pages yourself (not recommended)

Only when there is no API. Scraping boards stacks three problems: page structure changes, anti-bot measures (rate limits, captchas, IP bans), and the compliance call for how you use the data lands back on you.

Exchange disclosure pages (SSE / SZSE) are the most authoritative source, but they are pages for humans, not interfaces for programs — parsing is costly and field semantics are yours to align.

How to choose

Gotchas we hit in testing

• Every field value is a string: `"135160431.2"` must be `float()`-ed before math, or `+` concatenates
• Amounts are in CNY: 135160431.2 is 1.35 x 1e8 CNY — do not multiply by 10k again
• `instBuyRate` / `netBuyRate` are percentages (20 means 20%), not fractions
• `code` needs the market prefix (`sz301234` / `sh600664`); active-seat codes are semicolon-separated lists
• `data` is a Markdown string while `structured` is the array — use the latter in code
• The hot-money board (`hotmoney`) has no upstream data right now and returns "market data not found" (verified for 2026-09-17 through 09-22)

Common errors and boundaries

• `401` = missing key: this is a paid endpoint, anonymous calls are rejected
• `429` = rate limited: limits follow your key tier; slow down or upgrade
• `ok:false` = upstream fetch failed: we already failed over and it is not counted; one retry usually works
• Update cadence: after the close on trading days (no same-day board intraday — that is the disclosure schedule, not API latency)
• `date` accepts history (`YYYY-MM-DD`); empty means the latest trading day, and holidays will not turn into "today"