Filtration-Consistent Statistical Learning and Dependence-Aware Inference for Directional Forecasts of Iron Ore Futures
Year: 2026
Type: article
Abstract: We forecast the direction of iron ore futures returns on the Dalian Commodity Exchange using only information available at the forecast origin.Across 1,531 observations from 2020-2026, dependence-robust tests find no significant directional predictability for penalised logistic regression or three tree ensembles under daily-sign or triple-barrier labelling.Bootstrap upper confidence limits for the area under the ROC curve reach 0.54 and 0.60, respectively, leaving room for modest predictive effects.Changing only the predictor timestamp rule raises the realised Sharpe ratio from -0.06 to 0.66 for logistic regression and from -0.08 to 0.39 for the random forest.The formal tests yield the same conclusions under both timestamp rules.These results support reporting backtest performance alongside tests of predictive ability under an explicit informationavailability rule.
DOI: 10.62476/jcam171.24
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: bronze
OpenAlex ID: W7236801156
