Vector Autoregression in Forecasting COVID-19 Under-Reporting–Nepal as a Case Study

Abstract

This paper aims to understand and predict the dynamics of spread of COVID-19. It is based on government data on COVID-19 from February 1, 2021 to August 31, 2021. First, Vector Autoregression (VAR) model is used here to model the interrelationships between time series data of daily tested, infected, dead and discharged. The impact of under-reporting on interrelated variables is quantified. The behavior of the parameters of these VAR model is also analyzed. The entire time period of study is divided into three phases, according to the intensity of vaccination drive. The impact of vaccination in controlling the spread of the pandemic is measured by studying the behavior of the coefficients of VAR model for these three time periods. Then, Granger causality is also measured. At 10% level of significance, it is found that if the number of infected is under-reported today, this is due to the significant influence of number of infected until previous two days. The number of discharged one day ago and three days ago also significantly influence this number. Number of tests conducted two days ago also significantly contributes to this underreporting. The impact of latent variables on the spread of COVID-19 is measured here.