Algorithms for the evaluation of forecasts, filters and smoothers from a state-space model with the feature of time dependent dimension

Abstract

This thesis illustrates two approaches for the evaluation of forecasting, filtering and smoothing from a exible state-space model. Parameters of this model can be time dependent and the dimension of its state or observed vectors can vary over time. The first approach consists of establishing an algorithm based on the Kalman filter and Kalman smoother as well as properties derived from the model. Another approach is to reconstruct the model. In addition, an extension of the model is proposed.