Selecting single model in combination forecasting based on cointegration test and encompassing test

Area of Science:

• Econometrics
• Statistical Modeling
• Time Series Analysis

Background:

• Traditional combination forecasting methods often randomly select individual models.
• This approach overlooks the specific characteristics of the data being forecasted.
• Improved model selection is crucial for enhancing forecasting accuracy and reliability.

Purpose of the Study:

• To develop a principled approach for selecting single models in combination forecasting.
• To investigate the role of cointegration and encompassing tests in model selection.
• To provide guidance for optimizing the accuracy and stability of composite forecasts.

Main Methods:

• Utilizing cointegration tests to assess relationships between time series.
• Employing encompassing tests to evaluate the explanatory power of individual models.
• Conducting empirical analysis to validate the proposed selection strategy.

Main Results:

• A data-driven method for selecting single forecasting models was established.
• The study demonstrated that selecting a limited number of suitable models (<=5) enhances forecast accuracy.
• The proposed guidance leads to more stable and reliable composite forecasts.

Conclusions:

• Random model selection in combination forecasting is suboptimal.
• Cointegration and encompassing tests are valuable tools for selecting appropriate single models.
• A constrained selection of high-performing single models significantly improves forecasting accuracy and stability.