Formulação estrutural - abordagens clássica e bayeseiana: semelhanças e dessemelhanças

Abstract

Much use has been made recently of the analysis of a time series in terms of its non-observable components: trend, seasonal, cyclical and irregular. The structural model (as it is known in the literature) is one possible approach to the problem, in which the model is written in the form of state space, the corresponding state vector being made up of these non-observable components of the series. The Kalman filter can thus be used in sequential estimation of the state and forecasting of future observations. In this context can be found Harrison and Stevens procedure based on Bayesian statistics and Harvey's procedure based on classical statistics. This paper, a comparative study of these two approaches, seeks to point out the similarities and dissimilarities between them.