Reading A-share shareholder counts: chip concentration

1. Why look at shareholder counts (30 seconds)

Short-term moves are driven by money; medium-term moves are decided by chip structure. The holder count is the clearest public signal of that structure:

• Count falling → chips moving from retail to a few hands (big money / institutions) — usually read as accumulation
• Count rising → chips dispersing (more retail taking over) — usually read as distribution / fading
• Average shares held = total shares / count, inversely related, also measures concentration

Note: "usually read" is not a hard signal — combine with price level and who is on both sides

2. One request, three tables

`/v1/shareholder` is a paid endpoint. One parameter: `code`. The `data` is an object with `tables` (three tables).

import requests

BASE = "https://api.ashareapi.com/v1"
H = {"Authorization": "Bearer ct-your-key"}

b = requests.get(f"{BASE}/shareholder",
                 params={"code": "sh600667"}, headers=H, timeout=60).json()
d = b["data"]                      # note: data is an object, not an array

for t in d["tables"]:              # three tables, each with title/slug/rows
    print("== [%s] %s ==" % (t["slug"], t["title"]))
    for row in t["rows"][:5]:
        print("  ", row)

• `data` is an object (with `tables`), not a plain array — iterate with `data["tables"]`
• Do not use `body["structured"]` — shareholder data is a dict; that key is absent

3. What each of the three tables tells you

`holder_count` is the most valuable — it is reported per quarter and shows the count trend (not a single point).

4. Reading the fields

Two counts: `totalSHNum` (total) and `aSHNum` (A-share) are usually equal; `avgHoldShares` / `aAvgHoldShares` likewise.

5. Real case: Tai Ji Industry (600667) count triples in a year

Live holder_count table returned 2026-09-29:

• Trend: 153.6k → 178.2k → 548.9k, a 3.5x jump; avg shares fell from 13,614 to 3,810
• Meaning: lots of new retail flooded in, chips moved from concentrated to heavily dispersed — a classic "high-level distribution / retailisation" pattern
• Combine with price: if price rose then stalled, "count surge + price stalling" is a distribution risk; if it happens at a low after volume expansion, it may start a new rotation (needs more context; not a conclusion here)

Why this is a good case: it shows the proper use of the metric — read the trend, not a single point, and pair it with the price level.

6. Bulk scan (find "chips concentrating")

CODES = ["sh600667", "sh600519", "sz300750", "sh601318",
         "sh600036", "sz000858", "sh600030", "sh601899"]

for code in CODES:
    b = requests.get(f"{BASE}/shareholder", params={"code": code},
                     headers=H, timeout=60).json()
    d = b.get("data") or {}
    for t in d.get("tables", []):
        if t.get("slug") != "holder_count":
            continue
        rows = t.get("rows", [])           # latest first
        if len(rows) < 2:
            continue
        latest = int(rows[0].get("totalSHNum") or 0)
        prev = int(rows[1].get("totalSHNum") or 0)
        if prev > 0 and latest < prev * 0.9:     # count down >10% = concentrating
            print("%s  %s  count %d → %d  [concentrating]" % (
                code, rows[0].get("date"), prev, latest))

• Find concentration: `latest < prev * 0.9` (down >10% from prior) — chips moving to fewer hands
• Find dispersion (reverse): `latest > prev * 1.2` (up >20%) — retail crowding in (can be a avoid signal)
• Mind the lag: counts are quarterly (a 2026-06-30 count only appears by end of August) — not a "today" signal

7. Two common misreads (important)

• 1. Count data lags — quarterly, not real-time. Look at the `date` (as-of) field and do not treat it as the latest
• 2. Falling count is not always bullish, rising not always bearish — it is a mirror of chip structure. Combine with: price level (high-level dispersion is risky / low-level concentration may accumulate), who is on both sides (institution vs retail), and the overall market