EViews.com

Dear all,

This is the first time I'm trying to perform a Bayesian VAR in Eviews (Eviews 9) and I am a little bit confused. First of all, I have a database of 32 macroeconomic and financial variables. As I understood so far I have to run first a small VAR of 8 variables to find the right tightness for the large VAR (more specifically my paper says: "we use a Litterman prior, with tightness set so that the in-sample fit of the interest rate equation in the 32 variable VAR model is at the level achieved by a simpler 8 variable VAR". Does anyone have any idea of how I can do it??
Additionally, do I have to transform the timeseries to stationary or not?? Can I find the right lag order and forecast as I would do if I had a classical VAR??

Any help would be really appreciated.
Thank you!

When you run a BVAR, you have to set your priors. This can be done in several ways. The way chosen in your reference is to match an in-sample fit. But you can choose your priors other ways than this.

Usually, you transform the variables to a stationary VAR-representation before applying Bayesian estimation methods. But if you don't want to transform you variables, non-stationary BVAR:s are described in e.g. Villani (2009) Steady State Priors for Vector Autoregressions, Journal of Applied Econometrics, 24, 630-650, 2009:

http://onlinelibrary.wiley.com/doi/10.1 ... A3B.f04t02

Bayesian methods can be used to determine the lag structure, but for forecasting purposes a training-sample forecast evaluation can be done to determine the lag structure that works best for your purposes. Once you have estimated your model, you can use the Forecast procedure to do the forecasts.

/K

Dear KrilleJ,

Thank you for the prompt response.
I will use all the information you gave me. U were so helpful, however the authors of the article that I use for my research claim that for this research the most appropriate way to set the priors is through the in-sample fitting. Do u have any ideas of how I can do it??

Thanks again for your help.

The easiest way is to estimate the small VAR model and then use the hyper parameters of the Litterman prior to achieve the same in-sample fit (in whatever way it is measured) for the interest rate equation in the larger model.