Inferring COVID-19 spreading rates and potential change points for case number forecasts

Jonas Dehning; Johannes Zierenberg; Frank Paul Spitzner; Michael Wibral; Joao Pinheiro Neto; Michael Wilczek; Viola Priesemann

We chose to keep our main model comparatively simple, because of the small number of data points available during an epidemic outbreak. With such a low number of data points only a limited number of parameters can be effectively constrained. Hence, we chose to approximate a time-dependent spreading rate λ(t) by episodes of constant spreading rates λ i that are separated by three change points where a transition occurs. Our results show that this main model is currently sufficient for Germany: While we introduced fairly broad priors on the spreading rates, we obtained fairly narrow posterior distributions for each spreading rate λ i (Fig. 3) . We additionally evaluated obvious extensions of our main model with three change points by excluding weekend data that may suffer from under-reporting (Fig. S3) , and by using the more involved model that includes the four classes: susceptible, exposed, infected, recovered (Fig. S4 ). All of these models yield consistent results for the three change points, and all have LOO scores within one standard error of each other. Thus, we consider our main model to be sufficient for case numbers in Germany at present.