Obciążenie prognoz struktury uzyskanych przy pomocy łańcuchów Markowa

Abstract

The paper investigates a bias of forecasts of structure obtained from Markov chain models when the transition matrix of the chain is estimated using least squares methods and macrodata. The bias comes from a bias of the transition matrix estimators. Based on simulation methods it is shown that despite significant bias in the transition matrix estimates, bias of the forecasts is rather small. It is also acknowledged that analytical equation known from literature gives accu- rate estimates of the transition matrix bias, but it is not useful for assessing bias of the forecasts. The analysis is illustrated with forecasting of structure of Polish export by the main partners.