A multi-objective ensemble prediction model for interval-valued carbon price based on mixed-frequency data and sub-model selection

Accurate forecasting of carbon prices is crucial for climate policy design, investment decisions, and sustainable development. Existing literature predominantly emphasizes point-value predictions, while interval-value forecasting frameworks often ignore external influencing variables, particularly those derived from mixed-frequency data sources. Therefore, this paper introduces an interval-valued prediction model based on mixed-frequency data that combines modules such as frequency alignment, optimal sub-model choice, and multi-objective ensemble learning. First, factoring affecting international market price, energy sector dynamics, and domestic macroeconomics are selected as external influencing variables, and the frequency alignment module is utilized to transform high-frequency influencing variables into vectors under low-frequency observations. Subsequently, an effective sub-model selection algorithm identifies optimal sub-models for the sub-sequences with varying data characteristics. Lastly, a multi-objective ensemble module synthesizes the predictions of each reconstructed component, thereby yielding the ultimate interval-valued predictions for carbon prices. The findings of the experiments demonstrate that this model exhibits superior predictive accuracy compared to other benchmark models.