Reading main capital flow in A-shares

What one request gives you

import requests

KEY = "ct-your-key"
r = requests.get("https://api.ashareapi.com/v1/fund",
                 headers={"Authorization": "Bearer " + KEY},
                 params={"code": "sh600667"}, timeout=30)
d = r.json()["data"]

print(d["date"], d["close"])           # 2026-09-24 19.41
print("today  ", d["main_net"])        # -528559730.00
print("5-day  ", d["main_net_5d"])     # -571025057.00
print("10-day ", d["main_net_10d"])    #  531599448.00
print("20-day ", d["main_net_20d"])    #  -82259566.00
print("market rank", d["main_rank"])   # 5532

• The payload is an object (not an array): `main_net*` / `main_rank` / `lhb` / `lhb_details` / `block_trades` / `margin`
• Amounts are in CNY and arrive as strings (`"-528559730.00"`) — cast to float before comparing
• `main_rank` is the market-wide net-inflow rank (larger = further toward net outflow)
• The same call also returns top-trader boards, block trades and margin data — no extra endpoints needed

Read the four horizons together

One day lies. A single session is noise; the four horizons together carry the signal:

• Rank beats absolute value: 100M of inflow ranked #10 is nothing like ranked #3000
• `industry_rank` is the within-industry rank; in our measurement it returned `0`, meaning the figure was unavailable that day — not "ranked 0"

Top-trader boards: institutions or hot money?

`lhb` is the summary (reason / total buy / total sell / net buy); `lhb_details` is the seat-level detail (up to 10 rows) — the key to "who is trading".

for x in d["lhb"]:
    print(x["Reason"], "| buy", x["TotalBuy"], "| sell", x["TotalSell"], "| net", x["NetBuy"])
# measured: cumulative 20% deviation over 3 sessions | buy 2.177B | sell 2.570B | net -393M

for x in d["lhb_details"]:
    side = "buy" if x["Buy"] else "sell"
    print(f'{x["RankType"][:6]} #{x["Rank"]} {x["Name"]}  {side} {x["Buy"] or x["Sell"]}')
# measured (sh600667, 2026-09-24):
#   sell seat #5 UBS Shanghai Huayuanshiqiao Rd   sell 279M
#   sell seat #3 Goldman Sachs Shanghai Century Ave sell 487M
#   sell seat #2 Guotai Haitong HQ               sell 565M

• The seat name is the signal: known hot-money seats are short-term money; "institution" seats are funds
• Look at the net direction: when both sides appear, compare net buy, not "on the board = bullish"
• `HotMoneyTags` flags hot-money tags (empty string in our measurement = none)
• Being on the board is an outcome, not a cause: it means the stock already moved sharply; chasing then often means buying the top

Block trades and margin, in the same call

print("block trades", len(d["block_trades"]))
print("margin", len(d["margin"]))
# measured on sh600667: both are empty arrays [] -> no such data that day
# NOTE: an empty array is not a failure — just nothing on file for this stock

• Block-trade discount = someone sold below market (bearish short term); premium = someone paid up to buy
• Rising margin balance = leveraged longs (and a bigger stampede risk on a pullback)
• For dedicated views use `/v1/block-trade` and `/v1/margin-trade` (the latter supports batched codes)

Four common misreadings

• "Inflow means it will rise" — wrong. Flow is coincident or lagging: inflow with a flat price (accumulation) and outflow with a rising price (distribution) both happen
• Reading one day only — single-session noise is huge; look at least at the 5-day
• Treating rank as a verdict — rank is relative; in a bear market even #100 can be net selling
• Ignoring how "main" is defined — it usually means large/block orders, and thresholds differ by platform. We give a consistent multi-horizon series, so comparing across time and names is safer than raw absolutes

Boundaries (stated so you do not misuse it)

• No northbound flow intraday — there is no market-wide northbound endpoint
• No minute-level flow (we do not do minute data)
• No market-wide flow ranking table — `main_rank` is one stock’s rank among all stocks, not a list you can dump
• Key required — `/v1/fund` is a paid endpoint; the top-trader split `/v1/lhb` is paid too (type=institution/hotmoney/activeseat)
• Not investment advice — how to fetch and how to read, not what to buy