Estimation

 

By default SVAR uses a maximum of 12 lags for the lag order control. This should typically be sufficient for sensible VAR models. In the event that you want to change this maximum you can enter a new positive integer in this option field.

 

Whenever SVAR uses iterative estimators it needs to have a maximum number of iterations. By default this parameter is set to 5000. If you'd like to change this maximum you can do so here.

 

When SVAR uses iterative estimators it compares the output of a function, e.g. the maximum of the absolute value of the score vector, to a parameter called the convergence criterion. By default the software uses 1.0e-008. You can change this parameter in this edit field.

 

This option allows you to select which method SVAR should use when estimating VAR models. By default, a 3-step classical Maximum Likelihood technique due to Vlaar (2004) is used. It will estimate the cointegration relations and the reduced form of the VAR model using the Johansen (1996) ML technique. Given that you have selected to estimate a structural form VAR it will then estimate B, a matrix measuring the contemporaneous effects of the structural shocks, conditional on the reduced form and β. The second method is a 2-step classical Maximum Likelihood technique. It will estimate the cointegration relations and the reduced form of the VAR model using, again, Johansen's ML techniques. Given that you have selected to estimate a structural form, this method will use β from the first step and then estimate the reduced form parameters and B simultaneously. If you have chosen to only estimate a reduced form VAR model, then the 3-step and 2-step methods are equivalent. Note that if you've selected the option "Re-estimate beta when estimating alpha under restrictions" on the Cointegration tab, then the 3-step method is turned into a 2-step method (β and reduced form simultaneously, then B conditional on these estimates) and the 2-step method into a 1-step or "Full Information" method (β, reduced form and B simultaneously);  the text strings for the radio buttons will change accordingly. The third method uses Bayesian MCMC sampler(s) for estimating structural VAR's with cointegration restrictions; see Prior Distributions for additional information. Options for the Bayesian VAR analysis can be changed on the Bayesian tab.

 

By default SVAR "prefers" contemporaneous identification. This will then be the chosen identification on the "Identification & Estimation" dialog if there are such identifying restrictions available. From the default identification control you can change this behavior to either "Long-run" or "Contemporaneous & long-run" identification. Naturally, at run-time you can always change this once you reach the Identification & Estimation dialog.