Predicting Bitcoin Returns Using Artificial Neural Networks - An Application of Large Datasets to Convolutional Neural Networks and Long Short-Term Memory Based Artificial Neural Networks in Finance.

Abstract

Time series forecasting is one of the foremost challenges studied in finance. In this thesis various Convolutional Neural Network and Long Short Term Memory Artificial Neural Network models are used to predict Bitcoin returns. Previous literature has explored using data from Sentiment analysis of Social Media, and Blockchain information in isolation. This thesis seeks to combine the predictive power of earlier smaller models into a larger model that better utilizes a broader category of features in time series prediction. The resulting models are able to predict Bitcoin returns well, beating out simpler methods that do not utilize Artificial Neural Networks.