Inference on the dynamics of the COVID pandemic from observational data

Abstract

We describe a time dependent stochastic dynamic model in discrete time for the evolution of the COVID-19 pandemic in various states of USA. The proposed multi-compartment model is expressed through a system of difference equations that describe their temporal dynamics. Various compartments in our model are connected to the social distancing measures and diagnostic testing rates. A nonparametric estimation strategy is employed for obtaining estimates of interpretable temporally static and dynamic epidemiological rate parameters. The confidence bands of the parameters are obtained using a residual bootstrap procedure. A key feature of the methodology is its ability to estimate latent compartments such as the trajectory of the number of asymptomatic but infected individuals which are the key vectors of COVID-19 spread. The nature of the disease dynamics is further quantified by the proposed epidemiological markers, which use estimates of such key latent compartments.