This thesis proposes a new econometric methodology for the estimation and inference of macro- economic models in the presence of time variation in the parameters. A novel quasi-Bayesian local likelihood (QBLL) approach is established and it is shown that the method gives rise to as- ymptotically valid quasi-posterior distributions. In addition, in the special case of linear Gaussian models, expressions of the quasi-posteriors are derived in closed form, which simpli es inference and makes the use of MCMC unnecessary. Inference based on the QBLL approach, as a consequence of modelling parameter variation nonparametrically, is robust to di¤erent processes for the drifting parameters, as its validity does not depend on parametric restrictions typically imposed by alterna- tive state space models. In addition, the Bayesian treatment of the approach provides a remedy to the curse of dimensionality by accommodating large dimensional systems. We demonstrate that the proposed estimators exhibit good nite sample properties, and, unlike the alternative para- metric state space models, are robust to di¤erent parameter processes. We provide a variety of interesting macroeconomic applications and forecasting exercises to reduced-form VAR models. In addition, we develop the methodology to the estimation of structural DSGE models in the presence of parameter drift. We apply the proposed algorithms to di¤erent medium-sized DSGE models in order to study structural change in the parameters.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:765811 |
Date | January 2016 |
Creators | Petrova, Katerina |
Publisher | Queen Mary, University of London |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://qmro.qmul.ac.uk/xmlui/handle/123456789/23650 |
Page generated in 0.0018 seconds