A stock market trading system using deep neural network

Abstract

The stock market prediction is a lucrative field of interest with promising profit and covered with landmines for the unprecedented. The markets are complex, non-linear and chaotic in nature which poses huge difficulties to predict the prices accurately. In this paper, a stock trading system utilizing feed-forward deep neural network (DNN) to forecast index price of Singapore stock market using the FTSE Straits Time Index (STI) in t days ahead is proposed and tested through market simulations on historical daily prices. There are 40 input nodes of DNN which are the past 10 days’ opening, closing, minimum and maximum prices and consist of 3 hidden layers with 10 neurons per layer. The training algorithm used is stochastic gradient descent with back-propagation and is accelerated with multi-core processing. A trading system is proposed which utilizes the DNN forecasting results with defined entry and exit rules to enter a trade. DNN performance is evaluated using RMSE and MAPE. The overall trading system shows promising results with a profit factor of 18.67, 70.83% profitable trades and Sharpe ratio of 5.34 based on market simulation on test data.