Trading Strategy Evaluation
ML-based trading system using ensemble methods (BagLearner with Random Trees) to predict stock movements, with full backtesting simulation including transaction costs.
Overview
A comprehensive machine learning project for developing and evaluating automated trading strategies. The project implements ensemble learning using bagging with random decision trees, calculates technical indicators, and simulates market conditions with realistic transaction costs to develop profitable trading strategies.
Technologies
Key Features
- ✓BagLearner ensemble implementation from scratch
- ✓RTLearner (Random Tree) for trading signals
- ✓Technical indicators (SMA, Bollinger Bands, RSI, MACD)
- ✓Market simulation with transaction costs and slippage
- ✓Strategy backtesting with walk-forward analysis
- ✓Performance metrics (Sharpe ratio, cumulative returns, max drawdown)
Challenges
Designing effective feature engineering pipelines for financial data, implementing ensemble learning methods without external ML libraries, handling look-ahead bias in backtesting, and optimizing strategy parameters for out-of-sample performance.
Results
Bagged-Random-Tree learner (BagLearner over RTLearner) trained on 4 technical indicators (SMA, Bollinger Bands, RSI, MACD) materially outperformed a buy-and-hold benchmark in walk-forward backtests with realistic transaction costs and slippage.