Learning time series data using cross correlation and its application in bitcoin price prediction

Abstract

In this work, we developed an quantitative trading algorithm for bitcoin that is shown to be profitable. The algorithm establishes a framework that combines parametric variables and non-parametric variables in a logistical regression model, capturing information in both the static states and the evolution of states. The combination improves the performance of the strategy. In addition, we demonstrated that we can discovery curve similarity of time series using cross correlation and L2 distance. The similarity metrics can be efficiently computed using convolution and can help us learn from the past instance using an ensemble voting scheme.