Bayesian Time-varying Coefficients VAR Models

It is often difficult to justify the VAR assumption that model parameters are constant over time. For example, a basic VAR fitted to post-war macroeconomic data assumes that economic relationships have not changed since the mid-1940s. Two popular modeling approaches that do away with this assumption are the switching VAR and the time-varying coefficients VAR (TVCVAR). The switching VAR deals with occasional discrete changes (e.g., structural breaks), whereas the TVCVAR handles constant, smooth changes. The discussion here pertains to the latter.

The EViews implementation of TVCVAR uses a Bayesian framework. The Bayesian TVCVAR, or BTVCVAR, combines the TVCVAR model with a prior distribution. The BTVCVAR is popular even among those who do not identify as Bayesian because the prior provides a convenient way to induce shrinkage in a model that needs it.

Our discussion of BTVCVAR begins with an overview of methodology. We then demonstrate how estimation and other post-updating procedures are carried out in EViews. Next, we provide implementation details, and finish the discussion with an example.