255 Shades of Grey: Using Grey-Scaled Images and Convolutional Neural Nets to Forecast Tomorrow's Stock Returns

Abstract

Predicting stock returns has been a never ending endeavour of both, practitioners and academics. Accurate forecasts are crucial for investment decisions and performances as well as for analysing market microstructures. This paper offers an innovative approach towards forecasting based on Neural Nets and high frequency stock market data to predict the next day's stock return. The application uses intraday stock market data of the 30 Dow Jones constituents, which are transformed into grey-scaled images and fed into a Convolution Neural Net. The results reveal that the previous day images are a surprisingly accurate predictor of the next day return, implying that the information and patterns inhibited within images have remained undiscovered by standard approaches, such as autoregressive models.

Keywords: onvolutional Neural Networks, Forecasting, Intraday Data

JEL Classification: C45, C53, G17