Discovering the Trading Pattern of Financial Market Participants: Comparison of Two Co-Clustering Methods

Co-clustering is rapidly becoming a powerful data analysis technique in varied fields, such as gene expression analysis, data and web mining, and market baskets analysis. In this paper, two co-clustering methods based on smooth plaid model (SPM) and parallel factor decomposition with sparse latent factors (SLF-PARAFAC) are respectively applied to synthetic data set and investors' transaction-level data set from the China Financial Futures Exchange. We present the comparison between two methodologies. Both SLF-PARAFAC and SPM are efficient, robust, and well suited for discovering trading ecosystems in modern financial markets. We recognize temporal pattern differences of various trader types. The results help to develop a thorough understanding of trading behaviors, and to detect patterns and irregularities.