Algo-Trading System with ML & Automation

This project is a Python-based prototype of an algorithmic trading system. It automatically fetches stock data for NIFTY 50 tickers, applies a "buy the dip" trading strategy, backtests its performance, and uses machine learning to predict future price movements. All results, trades, and analytics are logged to Google Sheets, with real-time notifications sent via Telegram.

Key Features

• Automated Data Ingestion: Fetches daily stock data from Yahoo Finance with built-in resilience (retries and fallbacks).
• Rule-Based Trading Strategy: Implements a "buy the dip" strategy using RSI and Moving Average indicators.
• Performance Backtesting: Calculates detailed performance metrics, including P&L, win rate, and cumulative returns over a dynamic 6-month period.
• ML-Powered Predictions: Utilizes Decision Tree and Logistic Regression models to forecast next-day price direction, validated correctly with TimeSeriesSplit.
• Cloud Integration: Logs all trades, daily signals, and summary analytics to Google Sheets in separate, organized tabs.
• Real-Time Alerts: Sends start, completion, and error notifications to a Telegram chat.

Project Structure

The project is built with a modular architecture to ensure clean separation of concerns:

algo_trading_system/
├── __init__.py
├── main.py              # Main entry point to run the system
├── data_fetch.py        # Module for data ingestion from APIs
├── indicators.py        # Functions for technical indicator calculations
├── strategy.py          # Trading strategy logic and backtesting engine
├── ml_model.py          # Machine learning models and feature preparation
├── google_sheets.py     # Google Sheets integration and logging
└── telegram_notifier.py # Module for sending Telegram notifications

Setup & Execution

1. Prerequisites

• Python 3.8+
• A configured Python virtual environment (e.g., venv or conda).

2. Installation

1. Clone this repository

git clone https://github.com/your-username/AlgoTrading.git

2. Navigate to the project directory

cd AlgoTrading

3. Install all required packages

   pip install -r requirements.txt

4. Create a .env file in the project's root directory and add your credentials. See .env.example for the required format.

   TELEGRAM_BOT_TOKEN="your_telegram_bot_token"
   TELEGRAM_CHAT_ID="your_telegram_chat_id"
   GOOGLE_CREDENTIALS_JSON="secrets/service_account.json"
   GOOGLE_SHEET_NAME="YourGoogleSheetName"
   

5. Place your Google Cloud service_account.json file inside the /secrets directory.

3. Execution

To run the entire system, execute the main module from the project's root directory. The script will perform a 6-month backtest ending on the current day.

python -m algo_trading_system.main

Strategy Deep Dive

The core strategy is to identify a short-term buying opportunity within a confirmed long-term uptrend.

Buy Conditions

• RSI < 30: The stock is considered significantly oversold.
• 20-day SMA > 50-day SMA: The stock is in a confirmed bullish trend (a "Golden Cross" state).

Sell Conditions

A position is exited under two conditions to manage risk and lock in profits:

• RSI > 70: The stock is considered overbought.
• Death Cross: The 20-day SMA crosses below the 50-day SMA, signaling a potential trend reversal.

A Note on the Low Trade Count

The backtest results may show a low number of trades. This is an expected outcome and a feature of this highly selective strategy. The combination of an extremely oversold condition (RSI < 30) occurring within an established strong uptrend (20-DMA > 50-DMA) is inherently rare. This demonstrates that the system is working correctly by filtering for very specific, high-conviction setups as designed.

Machine Learning Analytics

The project includes a bonus ML component to forecast next-day price direction.

• Models: A Logistic Regression and a Decision Tree classifier.
• Features: A rich set of features is used, including RSI, MACD, various SMAs, volume indicators, volatility, and lagged values.
• Validation: The models are validated using TimeSeriesSplit cross-validation. This is crucial for financial data as it respects the chronological order and prevents lookahead bias, providing a realistic measure of performance. The resulting low accuracy scores (~50%) are an honest reflection of the difficulty in predicting financial markets with basic models.

Outputs & Deliverables

The system generates the following outputs:

• Console Logs: Real-time status updates and trade details.
• Google Sheets: Three organized tabs:
  1. Trades: A detailed log of every completed trade.
  2. Summary: High-level performance metrics per stock, including P&L, win rate, and ML accuracy.
  3. Signals: A daily log of all indicator values for debugging and analysis.
• Telegram Notifications: Alerts for the start and successful completion of the script, including a final summary.