A multi-factor dynamic time series measure for stock correlation analysis.

Area of Science:

• Quantitative Finance
• Computational Economics
• Data Science

Background:

• Traditional stock correlation analysis often fails to capture the multidimensionality of stock data and the dynamic time-lag effect (TLE).
• Existing similarity measures lack the sophistication to address the complex interplay of factors influencing stock behavior over time.

Purpose of the Study:

• To propose a novel Multi-Factor Dynamic Temporal Similarity Measure (MFDTSM) for more accurate stock correlation analysis.
• To address the limitations of existing methods in handling multidimensional stock data and the TLE in phase differences.

Main Methods:

• Developed an enhanced eXtreme Gradient Boosting (XGBoost) model integrated with Shapley Additive exPlanations (SHAP) to assess stock factor influence.
• Employed clustering of SHAP values for stock categorization and analysis of factor heterogeneity.
• Quantified TLE phase differences using cumulative distance matrices and optimal time series alignment paths.

Main Results:

• The MFDTSM method demonstrated improved accuracy in industry correlation (10%), linear correlation (16%), and stock correlation pricing (5%) compared to existing methods.
• Effectively categorized stocks and revealed heterogeneity in factor influence.
• Successfully quantified dynamic phase differences in TLE, enhancing similarity measure accuracy.

Conclusions:

• MFDTSM offers a significant advancement in analyzing complex stock market dynamics by incorporating multidimensional data and TLE.
• The method proves efficient and stable, outperforming existing techniques in various correlation analyses.
• Highlights the importance of considering dynamic temporal aspects and factor interactions for robust stock market insights.