With financial time series data, it makes more sense to select the first 80% of data for training and the last 20% for test because of its tendency to exhibit autocorrelation. correct me if i'm wrong.