Stock Price Prediction Using Convolutional Neural Networks on Various Time Frames

Abstract

The stock market is a place where stocks can be transferred, traded, and circulated. It has a history of 400 years and can be used as a channel for companies to raise funds. In this study, two Convolutional Neural Network (CNN) models are developed to forecast stock market prices, catering to distinct investment strategies: a short-term weekly model and a medium-term approximately three-month model. The weekly model utilizes a structure to predict Friday closing prices based on the daily closing prices from the preceding week, intending to capture the weekly trends. In contrast, the medium-term model is designed to comprehend the broader market movements by predicting the closing price on a day nearly twelve weeks ahead, using the past twenty days’ prices. Both models incorporate a Conv1D layer followed by batch normalization, pooling, and dense layers, calibrated to their respective temporal frameworks. The study’s findings demonstrate the challenges and intricacies of stock price prediction over varying time frames and propose enhancements for future models, including the incorporation of a wider array of market indicators and economic factors.