Forecasting the CoViD19 Diffusion in Italy and the Related Occupancy of Intensive Care Units

Livio Fenga

In order to extract valuable information from our data and, at the same time, decrease the total amount of uncertainty associated to the outcomes of the ARMA model, a resampling procedure has been employed. Among the several resampling methods for dependent data available -many of which freely and publicly available in the form of powerful routines working under software packages such as Python ® or R ® -the adopted resampling method is of the type Maximum Entropy Bootstrap (MEB). Proposed by Vinod (2006) and subsequently improved (see, e.g., Vinod (2016) ), it is based on basic assumptions which are different from those usually followed by standard schemes. In more details, while in the classic bootstrap an ensemble Ω represents the population of reference the observed time series is drawn from, in MEB a large number of ensembles (subsets), say {ω 1 , . . . , ω N } becomes the elements belonging to Ω, each of them containing a large number of replicates {x 1 , . . . , x J }.