The information on this page is based on a recursive delay model (SEPAR) for the spread of an epidemic proposed by Matthias Kreck (Bonn and Frankfurt) and Erhard Scholz (Wuppertal): SEPAR-model Arxiv version. The most striking application can be found in: Application.

The model can be used to describe the spread in the past as well as to make conditional predictions. For a while we thought that based on the available data of the past one can make a reasonable prediction. This is the case as long as the reproduction number is approximately constant. But in times where it continuously drops or increases over a longer period this must lead to unrealistic predictions. From now on we therefore publish only conditional predictions from for the number of actually infected as well as for the number of newly infected. Depending on the development of the last few days before the prediction we discuss different scenarios.

Scenarios lowering of the actual cases (last version Jan. 22, 2021)

Scenarios lowering of the newly infected (last version Jan. 22, 2021)

Explanation:

We give weakly forecasts typically on Thursdays or Fridays. In the past we observed longer intervals where the reproduction number is nearly constant, but also longer periods where it approximately linearly increases or decreases over longer times. We also observe that the slope in such phases is rather similar.

If during the last few days before a new forecast we don’t see signs of such a change we just apply our model under the assumption that this holds for the next weeks.

If we observe a change which might indicate a change of the constancy level we make conditional predictions assuming that the level is changing with a constant slope until it reaches a new constancy level which is a certain percentage of the last constancy level.

Such changes of the level might be the result of measures imposed by the government.