@article {Proskurinjfds.2021.1.065, author = {Oleksandr Proskurin}, title = {Does the CFTC Report Have Predictive Power: Machine Learning Approach}, elocation-id = {jfds.2021.1.065}, year = {2021}, doi = {10.3905/jfds.2021.1.065}, publisher = {Institutional Investor Journals Umbrella}, abstract = {This article investigates the forecasting power of features extracted from the Commodity Futures Trading Commission (CFTC) Commitments of Traders (COT) report for the most actively traded agricultural commodities: soft red winter wheat, corn, soybean, hard red winter wheat, soybean meal, and soybean oil. By comparing these data with existing CFTC report research papers, this article extracts advanced features from COT reports and applies a nonlinear machine learning approach to estimate the out-of-sample prediction power of the CFTC report.TOPICS: Commodities, information providers/credit ratings, big data/machine learning, performance measurementKey Findings▪ The COT report does not contain informative features predicting any of the agri-commodities if published on Friday with information from Tuesday.▪ Shifting report information from Friday to Tuesday significantly improves prediction power for wheat, corn, and soybean.▪ The performance of money managers and other (non-) reportable position dynamics are the most informative features in predicting the trend using the CFTC{\textquoteright}s COT report.}, issn = {2640-3943}, URL = {https://jfds.pm-research.com/content/early/2021/06/10/jfds.2021.1.065}, eprint = {https://jfds.pm-research.com/content/early/2021/06/10/jfds.2021.1.065.full.pdf}, journal = {The Journal of Financial Data Science} }