Probabilistic reconciliation of count time series

Forecast reconciliation is an important research topic. Yet, there currently neither formal framework nor practical method for the probabilistic of count time series. In this paper we propose a definition coherency and reconciled forecast which applies to both real-valued variables novel reconciliation. It based on generalization Bayes' rule it can reconcile real-value variables. When applied variables, yields probability mass function. Our experiments with temporal show major improvement compared Gaussian