This paper develops an econometric panel data model that can be used both to identify the dynamic effects of disease transmission factors and to forecast disease spread. The empirical model is derived from the canonical SIR epidemiological model of infectious disease spread. The model is estimated using near real-time, county-level data on mobility, weather, and COVID-19 cases. Both mobility and weather are found to have significant effects on COVID-19 effects up to 70 days ahead. Predicted values from the estimated model, augmented to incorporate recent vaccinations, provide out-of-sample forecasts of COVID-19 infections at the county and national levels. Prior forecasts are shown to have been fairly accurate, especially in terms of the geographical/cross-sectional distribution of COVID-19 infections and in terms of the national aggregate forecast. The latest forecasts, using data through February 19, 2021, predict steep declines in infections in most parts of the country over the next several weeks. Nationally, infections are predicted to fall by 59% over the subsequent 30 days. Decomposing the drivers of the latest forecast, the model indicates that accumulated natural immunity (i.e., cumulative infections to date, a.k.a. “seroprevalence”) is the primary factor exerting a strong downward pull on new infections.
J. Wilson, Daniel. 2020. “Weather, Mobility, and COVID-19: A Panel Local Projections Estimator for Understanding and Forecasting Infectious Disease Spread,” Federal Reserve Bank of San Francisco Working Paper 2020-23. Available at https://doi.org/10.24148/wp2020-23