How to Build a Northbound Capital Timing Strategy with Python and Tushare

The author, a data‑analysis enthusiast, shares personal finance reflections and introduces a timing strategy that increases regular investments when the market is in a prolonged trough and reduces them when the market is hot, using daily northbound capital data.

Northbound Capital?

In the Chinese stock market, "north" refers to mainland stocks and "south" to Hong Kong stocks. Northbound capital is the money flowing from Hong Kong (and international investors) into the mainland market, typically via the Stock Connect channels.

Historically, northbound capital outperforms most domestic investors, likely due to richer experience, broader information channels, or foreign speculative flows.

When monitoring daily trading, investors can track northbound buying; however, two concepts are often confused: net inflow and net buy. Net inflow includes both executed trades and unfilled orders, while net buy reflects only completed purchases.

Because net inflow is always greater than or equal to net buy, the author recommends focusing on net buy amounts for a more accurate view of actual foreign participation.

1. Import Libraries

# Import required libraries
import tushare as ts
import datetime
import pandas as pd
import numpy as np

2. Connect to Tushare with Token

# Replace the token with your own
token = 'replace_with_your_token'
pro = ts.pro_api(token)

3. Get Trading Calendar

# Get all trading days
trade_date = pro.trade_cal(start_date='20180101', end_date=datetime.datetime.today().strftime('%Y%m%d'))
date_list = list(trade_date[trade_date.is_open==1]['cal_date'].values)

4. Fetch Daily Northbound Capital Data

Because each request is limited to 300 rows, the data is retrieved in three batches and then concatenated.

# First batch
df_data1 = pro.moneyflow_hsgt(start_date=date_list[0:300][0], end_date=date_list[0:300][-1])
# Second batch
df_data2 = pro.moneyflow_hsgt(start_date=date_list[300:600][0], end_date=date_list[300:600][-1])
# Third batch
df_data3 = pro.moneyflow_hsgt(start_date=date_list[600:][0], end_date=date_list[600:][-1])
# Combine and sort
df_data = df_data1.append([df_data2, df_data3], ignore_index=True)
df_data = df_data.sort_values('trade_date', ascending=True).reset_index(drop=True)
# Rename columns for readability
df_data = df_data.rename(columns={'ggt_ss':'Shanghai', 'ggt_sz':'Shenzhen', 'hgt':'Shanghai Connect', 'sgt':'Shenzhen Connect', 'north_money':'Northbound', 'south_money':'Southbound'})

5. Unit Conversion

The original data is in millions; it is converted to billions for easier reading.

# Convert from million to billion
for col in ['Shanghai', 'Shenzhen', 'Shanghai Connect', 'Shenzhen Connect', 'Northbound', 'Southbound']:
    df_data[col] = df_data[col] * 0.01

6. Remove Non‑Trading Days

Some dates have no northbound activity due to holidays (e.g., Hong Kong SAR National Day). Rows where the Shanghai Connect column is NaN are filtered out.

# Filter out days without northbound trading
df_data2 = df_data.loc[~df_data['Shanghai Connect'].isna(), :].reset_index(drop=True)

7. Core Strategy Implementation

The strategy calculates a 252‑day rolling average and standard deviation of northbound net buy amounts, then defines upper and lower thresholds at ±1.5 × std. Signals are generated as follows:

signal = 'No Signal'
for index, row in df_data2.iterrows():
    if index < 252:
        continue
    recent = df_data2.iloc[index-252:index]
    avg = recent['Northbound'].sum() / 252
    std = recent['Northbound'].std()
    up_line = float(format(avg + std * 1.5, '.4f'))
    down_line = float(format(avg - std * 1.5, '.4f'))
    if row['Northbound'] >= up_line:
        signal = 'Bullish'
        print(f"{row['trade_date']}: <{signal}> Northbound net buy: {row['Northbound']:.4f}B, Upper: {up_line}B, Lower: {down_line}B")
    elif row['Northbound'] <= down_line:
        signal = 'Bearish'
        print(f"{row['trade_date']}: <{signal}> Northbound net buy: {row['Northbound']:.4f}B, Upper: {up_line}B, Lower: {down_line}B")
    if index == df_data2.shape[0]-1:
        print(f"
Latest data
{row['trade_date']}: <{signal}>
Northbound net buy: {row['Northbound']:.4f}B, Upper: {up_line}B, Lower: {down_line}B
")

The back‑test in the original report found that using an upper bound of +1.5 × std and a lower bound of –1.5 × std yielded the highest annualized return of 37.54 %.

Sample output (as of August 30) shows a continuous bearish signal since late July.

Note that the report’s original condition referred to "northbound capital inflow" while the Tushare API provides "net buy"; therefore, threshold values may need adjustment (e.g., using ±0.8 × std for net buy).

Readers are encouraged to experiment with different threshold parameters, back‑test the results, and adapt the strategy to their own dollar‑cost averaging plans.