Smooth multi-period forecasting with application to prediction of COVID-19 cases

Area of Science:

• Epidemiology
• Computational Statistics
• Time Series Analysis

Background:

• Forecasting methodologies are critical, gaining significant attention during the COVID-19 pandemic.
• Multi-period forecasting, predicting multiple future time points simultaneously, presents unique challenges.

Purpose of the Study:

• To develop and evaluate a novel approach for multi-period forecasting.
• To ensure predictions are 'smooth' across different forecast horizons.
• To apply the methodology to real-time distributed COVID-19 forecasting.

Main Methods:

• Proposed a novel forecasting approach enforcing smoothness across prediction horizons.
• Applied the method to point estimation using regression.
• Utilized quantile regression for interval prediction.

Main Results:

• Demonstrated the effectiveness of the 'smooth' prediction methodology.
• Successfully applied the technique to the CovidCast dataset.
• Validated the approach with a simulation example.

Conclusions:

• The proposed multi-period forecasting method provides a robust framework for time series prediction.
• The technique is particularly valuable for real-time epidemiological forecasting, such as for COVID-19.
• The 'smoothness' constraint improves prediction consistency across multiple horizons.