A Nonparametric Approach to Early Warning Signs using COVID-19 data from South Africa

Abstract

Introduction: Predicting outbreaks in epidemic events is crucial to developing and implementing effective mitigation measures by the public health sector.

Methods: This study examines the metric-based approach and the model-based approach (nonparametric Diff-Diffusion-Jump (DDJ) model) for early-warning signals (EWS) using the COVID-19 spread data from South Africa.

Results: The standard deviation is the most effective generic EWS. An increase in the standard deviation towards the critical points is shown with a positive Kendall-tau on all waves. The BDS test gives strong evidence of nonlinearity in all five waves. Conditional heteroskedasticity confirmed the periods of critical transitions that were captured by the generic EWS around day 140 and day 165 of wave 1, between day 110 and 120 of the second wave, and from day 120 onwards from wave 2. The results from the DDJ model are also consistent with results from the metric-based EWS. Sensitivity analysis for the robustness of the fitted model is based on the standard deviation and shows positive and increasing Kendall-tau estimates across all rolling windows. Results from these different approaches are consistent.

Conclusion: This study provides insights into pandemics that have similar characteristics as COVID-19 and informs policy making and implementation.