Shapley value-driven superior subset selection algorithm for carbon price interval forecast combination.

Area of Science:

• Econometrics
• Machine Learning
• Time Series Analysis

Background:

• Interval prediction accuracy depends on interval width and coverage, complicating model selection.
• Existing research often overlooks model selection challenges in interval forecast combination.

Purpose of the Study:

• To develop a robust model selection algorithm for interval forecast combinations.
• To improve the quality of interval forecasts by optimizing model subsets.

Main Methods:

• Introduced a model selection for interval forecast combination based on Shapley value (MSIFC-SV).
• Calculated Shapley values to assess marginal contributions and established a redundancy criterion based on interval scores.
• Iteratively removed redundant models to form an optimal subset for interval Bayesian weighting.

Main Results:

• MSIFC-SV significantly outperformed individual models and other subsets in carbon price forecasting.
• The method demonstrated superior performance across key metrics: prediction interval coverage probability (PICP), mean prediction interval width (MPIW), coverage width criterion (CWC), and interval score (IS).
• The approach was successfully validated on a housing price dataset, indicating its broad applicability.

Conclusions:

• MSIFC-SV offers a reliable method for selecting models in interval forecasting.
• The proposed approach generates high-quality interval forecasts with improved accuracy, width, and coverage.
• The universality of MSIFC-SV is confirmed through its successful application to diverse datasets.