Enhancing Stock Similarity Analysis with Phase-Embedded Multivariate Similarity Measure

Stock similarity analysis is essential for stock market prediction. However, existing measures of stock similarity often focus only on analyzing the magnitude differences of single stock feature, overlooking the impacts of multidimensionality, heterogeneity, and phase information of stock time series. To address this issue, this paper proposes a novel method for measuring similarity, namely PEMSM. It quantifies the contribution of multidimensional features using a combined approach of Random Forest and SHAP (SHapley Additive exPlanations), then analyzes feature heterogeneity using Coefficient of Variation Distance (CVD). Furthermore, PEMSM extracts phase information using the Hilbert transform and embedding it into the similarity calculation. Experiments demonstrate that compared to existing state-of-the-art methods, PEMSM improves the accuracy of linear correlation analysis and industry relationship analysis by 14% and 12%, respectively, and exhibits superior MAE and RMSE performance in prediction.