Bayesian

 

Select values for the hyperparameters governing the prior distribution for the short-run dynamics when the prior is informative. These parameters affect the prior variances for the parameters on lagged first differences of the endogenous variables, first differences of current and lagged exogenous I(1) variables, and current and lagged exogenous I(0) variables. A total of 4 parameters affect these variances, the so called baseline shrinkage (λ(b)), the lag length shrinkage (λ(l)), the off-diagonal/correlation shrinkage (λ(ρ)), and the exogeneity shrinkage (λ(e)). By default the values are given by λ(b)=0.3, λ(l)=1.0, λ(ρ)=0, and λ(e)=0.5. Details on how they affect the variances are given on the Prior Distributions page.

 

Select values for the hyperparameters determining the common standard deviation in the Gaussian distribution (conditional on Ω) for the parameters on the deterministic variables when the prior is informative (deterministic shrinkage). Positive values between 0.01 and 10000 are valid for the deterministic shrinkage hyperparameter. By default this hyperparameter is set to 1. Details on how it influences the prior distribution are found on the Prior Distributions page.

 

The alpha shrinkage hyperparameter determines the standard deviation of the Gaussian prior distribution for the α parameters conditional on β and Ω, where Ω is the covariance matrix for the reduced form residuals. This parameter is only effective when an informative prior has been chosen for alpha. By default this hyperparameters is equal to 10, but other positive values between 0 and 10000 can be selected. Details on how it influences the prior distribution are found on the Prior Distributions page.

 

For the covariance matrix of the reduced form residuals SVAR uses an informative prior by default. You can choose between using the informative and a non-informative prior for Omega. When an informative prior is selected, the prior distribution of Ω is inverted Wishart with parameters A and q. The matrix A can be selected to be equal to the maximum likelihood estimate of Ω in the full rank system (for any prespecified lag order - need not be the same as the model lag order), or a scalar times the identity matrix. In the former case, q = n+2 (where n is the dimension of Ω), while in the latter case q is at least n+2 and at most equal to the sample size. For details on how these priors are constructed and when the choice of a non-informative prior is used by SVAR see the Prior Distributions page.

 

The number of draws used to initialize the Markov Chain Monte Carlo (MCMC) sampler for simulating from the posterior distribution of the model parameters. The sampler is assumed to have converged at this number and draws to be saved are collected after it. The default value is 1,000 draws, but it's possible to skip this phase entirely by selecting 0 burn-in draws (not recommended). The maximum number of burn-in iterations is currently 10,000.

 

The number of draws from the posterior distribution of the model parameters that will be saved. The default value is 5,000 draws, but values ranging from 100 to 500,000 are also allowed.

 

Determines with which frequency SVAR saves draws from the posterior distribution. The default value is "1 in 1" meaning that all draws are saved, but any value in the save "1 in x" for a maximum x=20 are possible. Such a choice means that among x draws SVAR saves one, e.g. it saves draw 1, x+1, 2x+1, etc, until it has collected the full number of draws specified above. The total number of draws from the MCMC sampler is therefore "number of burn-in draws"+x"number of draws to save".

 

When drawing from the full conditional posterior distributions some draws may imply that the endogenous variables are explosive, i.e., the largest eigenvalue of the companion matrix has modulus greater than 1. By default SVAR will discard such draws, but this option makes it possible to save such draws.

 

A draw from the full conditional posterior distribution of a parameter group is considered bad if the covariance matrix for this parameter group is not positive definite. SVAR skips such bad draws by default by resetting the parameters to their last good draw and not saving the current. The maximum consecutive number of bad draws allowed by SVAR is 1000. By removing the check mark from this option, SVAR will instead abort the MCMC algorithm when a bad draw is located.