Forecasting Covid-19 in the United Kingdom: A dynamic SIRD model.

Abstract

Making use of a state space framework, we present a stochastic generalization of the SIRD model, where the mortality, infection, and underreporting rates change over time. A new format to the errors in the Susceptible-Infected-Recovered-Dead compartments is also presented, that permits reinfection. The estimated trajectories and (out-of-sample) forecasts of all these variables are presented with their confidence intervals. The model only uses as inputs the number of reported cases and deaths, and was applied for the UK from April, 2020 to Sep, 2021 (daily data). The estimated infection rate has shown a trajectory in waves very compatible with the emergence of new variants and adopted social measures. The estimated mortality rate has shown a significant descendant behaviour in 2021, which we attribute to the vaccination program, and the estimated underreporting rate has been considerably volatile, with a downward tendency, implying that, on average, more people are testing than in the beginning of the pandemic. The evolution of the proportions of the population divided into susceptible, infected, recovered and dead groups are also shown with their confidence intervals and forecast, along with an estimation of the amount of reinfection that, according to our model, has become quite significant in 2021. Finally, the estimated trajectory of the effective reproduction rate has proven to be very compatible with the real number of cases and deaths. Its forecasts with confident intervals are also presented.