Time series modelling in repeatedly conducted sample surveys

In this thesis, seemingly unrelated time series equation models in state space form are developed to model time series observed with repeatedly conducted probability samples. The models account for sudden changes in the input series and their variance due to changes in the survey process.

The standard Kalman filter used to analyse this type of time series models does not take into account the additional uncertainty created by estimating the unknown hyperparameters of the state space model with maximum likelihood. Bootstrap methods are applied to evaluate what the increase in variance of the Kalman filter estimates is if this additional uncertainty is taken into account. Finally, a comparison is made with a class of Bayesian multilevel time series models

Balabay, O. (2016). Time series modelling in repeatedly conducted sample surveys. Dissertation, Maastricht University, doi:10.26481/dis.20160511ob.