Comparative Study of Neural Networks and Decision Trees for Application in Trading Financial Futures

Abstract

The aim of this paper is to compare neural networks and decision trees in the ability to trade financial futures. Emphasis is made on correctness of comparison from practitioner's point of view. Contrary to other papers, we implemented actual trading strategy simulation with slippage and commission fees. Many research papers do not pay attention to correctness of experiments from practical point of view. Predictions are incorporated into trading algorithm and trading profits and Sharpe ratio are calculated. Experiments using 30 technical indicators and 45 different futures is repeated 16,895 times and evaluation is made using out of sample data. Results show that both methods have comparable accuracy and perform similarly.