Skip to main content
Fig. 3 | Financial Innovation

Fig. 3

From: Novel modelling strategies for high-frequency stock trading data

Fig. 3

The boxplots show F1 score improvements made by each of our proposed three modelling strategies, when they are used in SVM models (the top panel) and Elastic net models (the bottom model). Each box summarize F1 score improvements in 100 experiments conducted on different random subsets of the full data. A positive value in F1 score change indicates the strategy improve prediction performance. Hence, a box on the right hand side of the vertical dashed line (positioned at zero) indicate proposed strategy is helpful. To inference the significancy of improvement represented by each box, we calculate the raw p-value of Wilcoxon sign rank test, and applied the false discovery rate adjustment (Benjamini and Hochberg 1995) for multiple testing to avoid inflated Type-I error by multiple tests. The boxes corresponding to small adjusted p-values (less than 0.05) are colored in dark gray, which indicate a strategy significantly improve with prediction of that stock, whereas the light gray boxes represent no significant improvements

Back to article page