11. Putting It All Together: Case Studies in Feature Engineering for Trading Models

Pre-processing Data for Trading Models

Understanding Pre-processing:

• Critical Step: Pre-processing data ensures it is in the right form.
• Two Methods: Batch vs. Stream Processing.
  • Batch Processing: Efficient for predefined data chunks; ideal for personal trading models.
  • Stream Processing: Complex, real-time; essential for high-frequency trading.

Data Collection:

• Gather reliable data from sources like Yahoo Finance or Alpha Vantage.
• Use historical data to capture varying market conditions.
• Include additional indicators (e.g., trading volumes, financial statements) for context.

Data Cleaning:

• Remove missing values and outliers.
• Choose the timeframe based on trading strategy (e.g., daily for long-term).

Feature Engineering:

• Objective: Create "up" or "down" signals for stock predictions.
• Generate features such as:
  • Daily Returns: Percentage change in daily prices.
  • Moving Averages: Short-term (10 days) vs. long-term (50 days) for trend signals.
  • RSI: Indicator for overbought or oversold conditions.
  • Trading Volume: Assess the strength of movements.
  • Volatility: Identify potential trend changes.
  • Lagged Returns: Use past returns to forecast future prices.

Equip yourself with these insights to effectively build machine learning trading models.