An Empirical Algorithm for COVID-19 Nowcasting and Short-Term Forecast in Spain: A Kinematic Approach

Abstract

In the context of the COVID-19 pandemic, the use of forecasting techniques can play an advisory role in policymakers’ early implementation of non-pharmaceutical interventions (NPIs) in order to reduce SARS-CoV-2 transmission. In this article, we present a simple approach to even day and 14 day forecasts of the number of COVID-19 cases. The 14 day forecast can be taken as a proxy nowcast of infections that occur on the calculation day in question, if we assume the hypothesis that about two weeks elapse from the day a person is infected until the health authorities register it as a confirmed case. Our approach relies on polynomial regression between the dependent variable y (cumulative number of cases) and the independent variable x (time) and is modeled as a third-degree polynomial in x. The analogy between the pandemic spread and the kinematics of linear motion with variable acceleration is useful in assessing the rate and acceleration of spread. Our frame is applied to official data of the cumulative number of cases in Spain from 15 June until 17 October 2020. The epidemic curve of the cumulative number of cases adequately fits the cubic function for periods of up to two months with coefficients of determination R-squared greater than 0.97. The results obtained when testing the algorithm developed with the pandemic figures in Spain lead to short-term forecasts with relative errors of less than ±1.1% in the seven day predictions and less than ±4.0% in the 14 day predictions.

Keywords

• polynomial regression
• COVID-19 nowcasting and forecasting