Algorithmic Trading Strategy for the Colombo Stock Exchange (CSE)

This repository contains an algorithmic trading strategy for the Colombo Stock Exchange (CSE) using unsupervised learning. 📊

Performance: Our trading strategy outperformed an equal-weight ETF by 69.47% over a 5-year period.

Methodology: We selected the largest stocks by market capitalization in the CSE and applied K-Means clustering, optimizing a portfolio using the maximum Sharpe ratio. The aim was to potentially outperform market benchmark indices like an equal-weighted index of the top 150 CSE stocks. 📈

Key Findings:

Data Preprocessing and Feature Engineering

The initial dataset required significant preprocessing, including handling missing values and removing stocks with insufficient data. 📂 We also performed comprehensive feature engineering to incorporate valuable technical indicators, including:

• Garman-Klass Volatility
• RSI & MACD
• Bollinger Bands
• ATR
• Dollar Volume

Principal Component Analysis (PCA)

PCA revealed that RSI and ATR were the most influential features: 🔄

• RSI explained 27.55% of variance (PC1)
• ATR explained 17.01% of variance (PC2)

Together, they explained 44.56% of total variance.

Clustering Analysis

The optimal number of clusters (K=4) was determined using the elbow method and silhouette score. 💡 Clusters showed good separation when visualized using RSI and ATR as axes.

Trading Strategy Performance

The portfolio constructed using our clustering strategy consistently outperformed the equal-weight market index. 🚀 Key success factors include:

• Focus on stocks in cluster 3 (around RSI 70)
• Portfolio optimization with the maximum Sharpe ratio
• Dynamic monthly rebalancing
• Diversification constraints

Limitations and Considerations

The strategy relies heavily on technical indicators. Transaction costs and market impact weren’t considered. ⚠️