Bayesian VAR (BVAR) Models and Implementation in EViews скачать в хорошем качестве

Bayesian VAR (BVAR) Models and Implementation in EViews

The Bayesian Vector Autoregression (BVAR) model provides a highly effective solution to the severe "over-fitting" problem commonly encountered in traditional VAR models due to the large number of estimated parameters. Rather than relying solely on sample data, BVAR systematically incorporates non-data prior information (prior distributions) to induce coefficient shrinkage toward a simpler, stylized baseline model. This Bayesian approach significantly improves the precision and reliability of out-of-sample forecasts and impulse response functions (IRFs). Popular prior selections supported natively include the Litterman/Minnesota, Normal-Wishart, Sims-Zha, and Giannone-Lenza-Primiceri (GLP) priors. To estimate a BVAR model in the EViews software, select Quick - Estimate VAR... from the main menu or type var in the command window. In the VAR Specification dialog, select the Bayesian VAR radio button under the VAR Type section. You first specify your endogenous variables, exogenous variables, and lag intervals on the Basics tab exactly like a standard VAR. Next, EViews provides three specific tabs for BVARs: the Prior type tab lets you select your desired prior and initial residual covariance calculation options; the Hyper-parameters tab allows you to fine-tune scalar hyper-parameters (such as overall prior tightness); and the Options tab configures optimization or sampling algorithms (like the Gibbs sampler). Upon clicking OK, EViews generates the posterior estimates, allowing you to seamlessly perform Bayesian forecasting and Bayesian impulse response analysis to rigorously evaluate dynamic uncertainties.