This project provides a Python tool for backtesting and selecting the best-performing strategies to spot market time series, applying only machine learning algorithms.

As main advantages, the project provides:

• uses supervised machine learning to generate trading signals.
• recurring trading signals via Telegram channel.
• open-source code, allowing flexibility in specifying machine learning models and comparing all strategies.

Telegram open channel with daily signals run via GitHub Actions. Anyone can sign up to get a feel for what this bot can offer. t.me/market_forecasting_public

• Data download: Downloads market data through the Yahoo Finance API.
• Machine-learning-based indicators: Implements machine-learning-based strategies to identify possible trends.
• Strategy backtesting: Runs historical backtests, generating figures and summaries for decision-making.
• Performance evaluation: Assesses results using a weighted objective function and ranks the best strategies.
• Predict future prices: Make predictions based on supervised machine learning, applying applying statistical models or non-parametric algorithmic models.
• Configuration files: Uses .env for private environment variables, .json for tickers list, .jsonfor indicators list, and .csv for strategies list.

The project currently supports supervisioned machine learning strategies for generating trading signals, using the following methods:

• Linear Regression
• Decision Trees Regressor
• Random Forest Regressor
• Gradient Boosting Regressor
• Extra Trees Regressor
• K Neighbors Regressor
• Autoregressive Integrated Moving Average (ARIMA) models

These options allow the user to compare the performance of different approaches within the selected market.

1. Install dependencies:

    pip install pandas
    pip install numpy
    pip install yfinance
    pip install requests
    pip install python-dotenv
    pip install scikit-learn
    pip install statsmodels

2. Configure tickers and indicators

   • In config.json add the various configuration parameters.
   • In tickers.json add the stock codes to analyze.
   • In indicators.json add the indicators to generate.
   • In strategies.csv list the stocks to generate trading signals, each with its corresponding best strategy.

3. Configure Telegram

   • Create a Telegram bot and obtain its TOKEN.
   • Create a Telegram channel and obtain its CHAT_ID.
   • Add the bot as channel administrator.
   • Add keys to .env file to be read by market_forecaster.py.

4. Run the script

   • To run the batch of backtests, execute:

     python market_forecaster.py

• Backtest chart with Random Forest

  After running market_forecaster.py it generates strategy charts, spreadsheets for each ticker, and a summary with results sorted by best. The generated figures follow the example below:

  

  Notice that the asset ends the evaluated period near its initial price, so a Buy & Hold strategy would yield approximately a zero return. On the other hand, strictly following the machine learning strategy with Random Forest Regressor would produce above 160% return over the same period, excluding any trasactions fees. Furthermore, short selling operations are being considered by default in calculations, though they can easily be disabled in the backtest.

• market_forecaster.py → Main file for backtesting and selecting the best strategies.
• tickers.json → List of tickers to analyze.
• indicators.json → List of machine-learning-based indicators to test.
• strategies.csv → List of selected strategies for trading signals, including tickers and their indicators.

⚠️ We are not responsible for any financial losses resulting from the use of the strategies or signals generated by this code.

• Contributions are welcome! Open an issue or submit a pull request.
• Future improvements and new features may be added, including:
  • more machine learning methods; ✅
  • statistical methods (ARIMA); ✅
  • improve objective function with new weights and presets;
  • use optimizer for the objective function.

This repository is independently maintained, only in free time. If you find the code useful and wish to support its continued development, consider donating:

• PayPal

Your support helps keep the project alive and evolving, by adding new indicators, improvements, and documentation.