Deep Learning based Stock Trading Algorithms

Abstract

Nowadays, machine learning is the hottest topic in the world. Stock price prediction is a challenging task of machine learning. Since the stock market is a likely perfect competitive market, providing lots of data and information to the public, it means the stock market is one of the best places to validate the performance of machine learning. According to Robert J. Shiller's opinion, the stock market value should reflect the actual value of the company in long-term period. Is it true? The purpose of this project is applying DNN (Deep Neural Network), CNN (Convolutional Neural Network), SVM (Support Vector Machine) according to two years quarter financial reports and stock market information to predict 100-days-SMA (simple moving average) price after 6 months. Profit can be yield by apply automatic trading algorithms using the prediction from the combined model by DNN, CNN and SVM. Experiments used stocks in Dow Jones Industrial Average to prove that stock price is related to company financial status and stock market information. It can predict the trend of stock price and achieve about 90% accuracy in selecting increased stocks, applying trading algorithms and averagely better than the real situation. The experiment performance demonstrated the stock market value is predictable by combining financial status and stock market information on designed models.