Abstract: We use the recent development in deep learning technology to forecast stock prices. Focusing on image-type big data, we predict future stock prices using a convolutional neural network (CNN) model trained by visual representations of stock price data and technical indicators. We find that including technical indicators partially increases accuracy. The model with an input range of five days is the most accurate but is likely to be not appropriately learned, considering the recall, precision, and test datasets. On the contrary, training the model using past 20-day images along with technical indicators results in the greatest difference between the precision and label means of the test dataset.

Keywords: Convolution neural networks; Stock chart image; Stock price forecasting; Technical indicators

JEL codes: G11, G12, G17

DOI: ...