Modelling the Spread of COVID-19 in New York City

This paper proposes a methodology to predict the increase in the number of confirmed COVID-19 cases in the city of New York at the zip code level. We concentrate on the initial period of the pandemic spanning from March 31 to June 16, 2020. To do this, we propose a Poisson regression model for count data that includes a large set of covariates reflecting socioeconomic conditions at neighbourhood level and spatial effects. The sensitivity of the predictions of the number of cases to the specific choice of the regressors is controlled for by also considering an emsemble prediction model given by Bayesian model averaging. Our results extend related studies by showing that variables such as population size, its share of the elderly, the self-employment rate, income per capita, and the percentage of workers in the educational and healthcare sectors not only explain the cross-sectional variability in the number of new confirmed cases but also have out-of-sample predictive ability. Our pointwise forecasts display reasonable mean square prediction errors and the associated interval forecasts accurate empirical coverage probabilities suggesting the suitability of the methodology for prediction of the number of infections.