Nonlinear models: a case of the COVID-19 confirmed rates in top 8 worst affected countries

Abstract

Over the last 9 months, the most prominent global health threat has been COVID-19. It first appeared in Wuhan, China, and then rapidly spread throughout the world. Since no treatment or preventative strategy has been identified until this time, millions of people across the world have been seriously affected by COVID-19. The modelling and prediction of confirmed COVID-19 cases have been given much attention by government policymakers for the purpose of combating it more effectively. For this purpose, the modelling and prediction performances of the linear model (LM), generalized additive model(GAM) and the time-varying linear model (Tv-LM) via Kalman filter are compared. This has never yet been undertaken in the literature. This comparative analysis also evaluates the linear relationship between the confirmed cases of COVID-19 in individual countries with the world. The analysis is implemented using daily COVID-19 confirmed rates of the top 8 most heavily affected countries and that of the world between 11 March and 21 December 2020 and 14-day forward predictions. The empirical findings show that the Tv-LM outperforms others in terms of model fit and predictability, suggesting that the relationship between each country’s rates with the world’s should be locally linear, not globally linear.