Estimate the Parameters of the Weighted Exponential Regression Model for Panel Data

Abstract

The parameters of the weighted exponential regression model for panel data are estimated using the maximum likelihood method which represents the aim of this paper. Weekly infection and recovery ratios of COVID-19 data are predicted, where the model is converted from its nonlinear into a linear state using the Taylor series. The novelty of this paper lies in dealing with nonlinear panel data. Furthermore, the panel data of the model are tested to determine whether the data follows fixed or random effects by the Hausman test, as well as the exclusion of the pooled effects because the model does not include the intercept term. The simulation is depended on the generated data to compare the fixed and random effects models for different sample sizes (5, 10, 20, 30). COVID-19 data is used for three Iraqi governorates to represent the panel data model. Three months ,May, June, and July of 2022 are taken to represent the research sample and then predict the ratios of infection and recovery for the next three months. Depending on the Bayesian Information Criterion (BIC) and Akaike's Information Criterion (AIC), the random effects of the weighted exponential regression give better results than the fixed effects. Depending on this, we predict the weekly infection ratios of COVID-19 in Iraq that will decrease during the next ten weeks.