Improving the forecasting performance of temporal hierarchies.

Area of Science:

• Time series analysis
• Forecasting methodology
• Statistical modeling

Background:

• Temporal hierarchies are effective for multi-horizon forecasting but face limitations tied to base forecast methods and data characteristics.
• Existing methods often struggle with accuracy and bias, especially when dealing with seasonal data patterns.

Purpose of the Study:

• To address limitations in temporal hierarchy forecasting.
• To introduce and evaluate strategies for improving forecast accuracy and reducing bias.
• To enhance the reliability of forecasts across various planning horizons.

Main Methods:

• Combining forecasts from multiple forecasting methods.
• Implementing bias adjustment techniques on base forecasts.
• Selective application of temporal hierarchies to mitigate seasonal shrinkage effects.
• Evaluation using monthly data from the M and M3 competitions.

Main Results:

• The proposed strategies demonstrate significant potential for improving temporal hierarchy performance.
• Improvements were observed in both forecast accuracy and bias reduction.
• The strategies can be applied independently or in combination with existing reconciliation methods.

Conclusions:

• The investigated strategies offer a promising approach to overcome current limitations in temporal hierarchy forecasting.
• These methods enhance the overall effectiveness and reliability of hierarchical forecasting systems.
• Further application of these techniques can lead to more robust and accurate multi-horizon forecasts.