The current COVID-19 wave will likely be mitigated in the second-line European countries

Samuel Soubeyrand; Melina Ribaud; Virgile Baudrot; Denis Allard; Denys Pommeret; Lionel Roques

We used a linear interpolation in Equation (2) because of its simplicity. The use of splines could be investigated to evaluate whether this refinement improves forecast performance. Forecast performance could also be improved by: 1. optimizing the smoothing parameter or even by changing the form of the smoother that we used (e.g. local polynomial smoother, kernel smoother with a continuous kernel, etc.); 2. refining the penalization and the weighting introduced in Equation (4); 3. transforming the mixture model into a regression model and applying, for example, an elastic net penalization to reduce the number of likely predictors; 4. incorporating multiple-orders auto-regressive terms in the model of Y 0 to better account for temporal dependencies, but the auto-regressive terms should certainly be modified depending on the stage of the epidemics (increase stage, decrease stage and plateau); 5. incorporating additional covariates to account for temporal characteristics that are specific to the focal country (e.g., a change in the national control strategy).