Reproducibility Package for "Horizon-dependent forecast performance under structural change: a rolling-origin benchmark for global COVID-19 incidence

This repository contains the reproducibility materials for the study “Horizon-dependent forecast performance under structural change: a rolling-origin benchmark for global COVID-19 incidence.” The study evaluates short-horizon forecasts of global daily COVID-19 incidence using a rolling-origin framework over the period 22 January to 27 July 2020. Forecast horizons of 1, 3, 7, and 14 days are considered. The benchmark includes Naive, Seasonal Naive, Drift, ARIMA, ETS, Prophet, XGBoost, and LSTM models. The repository contains processed analytical data, derived target variables, and analysis materials supporting: deterministic forecast evaluation using MAE, RMSE, sMAPE, and MASE; retrospective regime-wise analysis under structural change; Diebold–Mariano forecast-comparison tests; probabilistic forecasting with ARIMA, ETS, and Prophet using empirical coverage, interval width, weighted interval score (WIS), and continuous ranked probability score (CRPS); robustness analyses for alternative segmentation settings, training-window policies, reporting-coverage thresholds, and target construction. The original COVID-19 incidence data were obtained from the publicly available Corona Virus Report dataset hosted on Kaggle and based on data from the Johns Hopkins University Center for Systems Science and Engineering (JHU CSSE).