Forecasting credit portfolio components with a Markov chain model

We consider the forecasting problem for components of a bank’s credit portfolio, in particular, for the share of non‐performing loans. We assume that changes in the portfolio are described by a Markov random process with discrete time and finite number of states. By the state of a loan we mean that it belongs to a certain group of loans with respect to the existence and duration of arrears. We assume that the matrix of transitional probabilities is not known exactly, and information about it is collected during the system’s operation.