A dynamic ensemble approach for multi-step price prediction: Empirical evidence from crude oil and shipping market

Price forecasting is critical for business management decision making and planning. However, accurate price predicting faces daunting challenges due to data drift and performance degradation. To this end, we propose a dynamic ensemble prediction method by introducing a "metabolic mechanism" and building an effective investment decision support platform. We generated ten individual models and provided eight ensemble strategies under static and dynamic scenarios to test and verify the model's prediction performance. The results reveal that the dynamic model presents excellent performance, obtaining the slightest prediction error and the most negligible fluctuation risk. Moreover, under the same ensemble strategies, the dominance of the dynamic model was verified by the improvement rate. The dynamic model can deliver more successful trading signals, demonstrating its importance in investment decisions. Thus, the proposed model benefits practitioners and plays a significant supporting role in investment decision making.