Biclustering Learning of Trading Rules

Area of Science:

• Quantitative Finance
• Machine Learning
• Data Mining

Background:

• Technical indicators are crucial for financial trading decisions, but identifying effective patterns is challenging.
• Existing methods struggle to extract actionable trading rules from complex historical financial data.

Purpose of the Study:

• To propose an innovative biclustering mining approach for discovering effective technical trading patterns.
• To develop a novel trading system integrating biclustering and K-nearest neighbor (K-NN) classification.

Main Methods:

• Applied biclustering algorithms to historical financial data to identify combined indicator patterns.
• Utilized a modified K-nearest neighbor (K-NN) method for classifying trading days into buy, sell, or no-action signals.
• Implemented the biclustering algorithm and the K nearest neighbor (BIC-K-NN) system on four historical datasets.

Main Results:

• The BIC-K-NN system successfully identified effective technical trading patterns and rules.
• Experimental results showed the proposed trading system significantly outperformed the buy-and-hold strategy.
• The BIC-K-NN system demonstrated superior average performance compared to three previously reported intelligent trading systems.

Conclusions:

• The proposed BIC-K-NN method offers an effective approach for discovering trading rules from historical financial data.
• This biclustering-based trading system shows promise for investment applications across various financial markets.
• The study highlights the potential of biclustering algorithms in financial market analysis and automated trading.