Model uncertainty and Bayesian model averaging in vector autoregressive processes
Year: 2006
Type: preprint
Abstract: Economic forecasts and policy decisions are often informed by empiri-\ncal analysis based on econometric models. However, inference based upon\na single model, when several viable models exist, limits its usefulness. Tak-\ning account of model uncertainty, a Bayesian model averaging procedure is\npresented which allows for unconditional inference within the class of vector\nautoregressive (VAR) processes. Several features of VAR process are investi-\ngated. Measures on manifolds are employed in order to elicit uniform priors\non subspaces de ned by particular structural features of VARs. The features\nconsidered are the number and form of the equilibrium economic relations\nand deterministic processes. Posterior probabilities of these features are used\nin a model averaging approach for forecasting and impulse response analysis.\nThe methods are applied to investigate stability of the Great Ratios in\nU.S. consumption, investment and income, and the presence and e¤ects of\npermanent shocks in these series. The results obtained indicate the feasibility\nof the proposed method.
Authors: Rodney W. Strachan, Herman K. van Dijk
Institutions: University of Leicester, Erasmus University Rotterdam
Language: English
Domain: Social Sciences
Open Access status: green
Funder: University of Leicester
OpenAlex ID: W2148552711
