Using Linear Regession to Trade

Using Linear Regession to Trade | Generated by AI

Home PDF

Great question! You’re combining two tasks here:

Fetching historical stock data (e.g., Tencent) using TigerOpen API
Using that data with your existing numpy-based linear regression code

Will it work as-is?

No, not directly. Your current ML code expects x_train and y_train as numpy arrays of numeric values representing features and labels (e.g., population vs. profit). But the TigerOpen API returns historical stock bars (OHLCV data), which is a time series with multiple fields per timestamp.

You need to:

Properly extract and preprocess the stock data into suitable x_train and y_train arrays.
Adapt your model and code to the stock data context (e.g., predict closing price from previous days).

How to adapt your stock data for your linear regression model?

Step 1: Fetch historical data for Tencent (symbol `'00700'`)

bars_dict = quote_client.get_bars(
    symbols=['00700'],
    period=BarPeriod.DAY,
    begin_time='2024-01-01',
    end_time=datetime.now().strftime('%Y-%m-%d'),
    limit=100,
    right=QuoteRight.BR
)

bars = bars_dict.get('00700', [])

Step 2: Extract features and labels

A simple approach:

Use the day index as x_train (e.g., 0, 1, 2, …)
Use the closing price as y_train

import numpy as np

x_train = np.array([i for i in range(len(bars))])
y_train = np.array([bar.close for bar in bars])

Alternatively, you can use previous day’s close as feature to predict next day’s close, or more sophisticated features.

Step 3: Use your linear regression code

Now your x_train and y_train are numpy arrays of numbers, so your code should work fine after you fix your compute_cost and compute_gradient functions (which currently return zero or zero gradients).

Example: Implement compute_cost and compute_gradient

def compute_cost(x, y, w, b):
    m = x.shape[0]
    total_cost = 0.5 * np.sum((w * x + b - y) ** 2) / m
    return total_cost

def compute_gradient(x, y, w, b):
    m = x.shape[0]
    error = w * x + b - y
    dj_dw = np.sum(error * x) / m
    dj_db = np.sum(error) / m
    return dj_dw, dj_db

Summary

Fetch historical stock data with TigerOpen API
Convert the time series data into numeric arrays (x_train, y_train)
Implement or fix your cost and gradient functions
Then your gradient descent and plotting code will work to fit a linear model to stock price data

Bonus: Plotting stock prices

Your scatter plot will show the closing price over days, and the regression line will fit a trend.

If you want, I can help you write a complete example integrating TigerOpen data fetching with your numpy linear regression code! Just ask.

Back 2025.04.20 Donate