Short-term forecast of U.S. COVID mortality using excess deaths and vector autoregression

Abstract

We analyze overall mortality in the U.S. as a whole and several states in particular in order to make conclusions regarding timing and strength of COVID pandemic effect from an actuarial risk analysis perspective. No effort is made to analyze biological or medical characteristics of the pandemic. We use open data provided by CDC, U.S. state governments and Johns Hopkins University. In the first part of the paper, we suggest time series analysis (ARIMA) for weekly excess U.S. mortality in 2020 as compared to several previous years’ experience in order to build a statistical model and provide short-term forecast based exclusively on historical mortality data. In the second half of the paper, we also analyze weekly COVID cases, hospitalizations and deaths in 2020 and 2021. Two midwestern states, Minnesota and Wisconsin, along with geographically diverse Colorado and Georgia, are used to illustrate global and local patterns in the COVID pandemic data. We suggest vector autoregression (VAR) as a method of simultaneous explanatory and predictive analysis of several variables. VAR is a popular tool in econometrics and financial analysis, but it is less common in problems of risk management related to mortality analysis in epidemiology and actuarial practice. Efficiency of short-term forecast is illustrated by observing the effect of vaccination on COVID development in the state of Minnesota in 2021.