Predicting Short-Term Price Changes in Cryptocurrencies Using Limit Order Book Data, Support Vector Machines, and Temporal Convolutional Networks

Abstract

As cryptocurrencies enter the mainstream and institutional traders become more involved, one would expect alpha generation to become more difficult as markets become more efficient. However, even if the trading of cryptocurrencies has become widespread among the general population and some firms, this new asset class is far from the efficiency we see in other markets like equities. The goal of this research is to examine the efficiency of cryptocurrency markets by developing trading strategies and to examine the efficacy of using machine learning techniques on cryptocurrency limit order books (and optionally, trade data) to predict short-term price movements. Specifically, this research will explore intra-day trends in cryptocurrency markets to find evidence of human traders, and support vector machines and temporal convolutional networks will be used to learn features in the limit order book and trade stream data that may predict future mid-price changes.