Time-Series K-means in Causal Inference and Mechanism Clustering for Financial Data

This paper investigates the application of Time Series K-means (TS-K-means) within the context of causal inference and mechanism clustering of financial time series data. Traditional clustering approaches like K-means often rely on static distance metrics, such as Euclidean distance, which inadequately capture the temporal dependencies intrinsic to financial returns. By incorporating Dynamic Time Warping (DTW) as a distance metric, TS-K-means addresses this limitation, improving the robustness of clustering in time-dependent financial data. This study extends the Additive Noise Model Mixture Model (ANM-MM) framework by integrating TS-K-means, facilitating more accurate causal inference and mechanism clustering. The approach is validated through simulations and applied to real-world financial data, demonstrating its effectiveness in enhancing the analysis of complex financial time series, particularly in identifying causal relationships and clustering data based on underlying generative mechanisms. The results show that TS-K-means outperforms traditional K-means, especially with smaller datasets, while maintaining robust causal direction detection as the dataset size changes.