Purpose: The objective of this research is to present a comprehensive framework for identifying anomalies in Commitments of Traders (COT) report data and to investigate their role in detecting economic trends, market disruptions, and sudden changes.
Methodology: In this study, following the preprocessing of the COT data, statistical methods, including the standard score and the interquartile range, were combined with machine learning algorithms, including Isolation Forest and One-Class Support Vector Machine. Then, by creating ensemble anomalies, the points identified as anomalous by all methods were extracted. Subsequently, linear and non-linear dimensionality reduction methods PCA, Isomap, UMAP, and LLE, were applied to the COT dataset, and the One-Class Support Vector Machine, Local Outlier Factor, Isolation Forest, and K-Means algorithms were implemented for anomaly detection.
Findings: The findings demonstrated that the detected anomalies closely coincide with prominent macroeconomic disruptions, notably the 2008 Global Financial Crisis and the economic shock of the 2020 COVID-19 pandemic. Additionally, the consensus framework, grounded in the intersection of model outputs, effectively filtered alarms driven by algorithm-specific sensitivities, thereby yielding a focused subset of 1,638 observations characterized by the highest inter-algorithm consensus.
Originality/Value: This research, by integrating statistical methods, machine learning, and Manifold Learning, provides an accurate and reliable framework for analyzing complex financial market data and creates the foundation for developing interactive dashboards for real-time market monitoring.