Exploring the Use of Data at Multiple Granularity Levels in Machine Learning-Based Stock Trading