First order terms in third order approximation misbehave


We are working with a third order approximation of a model. We have found that the IRFs are highly unstable. When investigating further, we found that they are unstable even if we simulate the model using only the first order terms. This is puzzling because the model is well behaved when doing a first order approximation. We compared oo_.dr.ghx and oo_.dr.ghu that we get from setting order = 1 to the oo_.ghx and oo_.dr.ghu that we get from setting order = 3, and they are very different, which is very odd. Could this be accounting for the unstable behavior? If yes, what can we do? Is it “kosher” to use the matrices from order=1 mixed up with the higher order ones calculated in order=3?

Thanks a lot

The linear model must always be stable as it has to satisfy the BK-conditions. Hence, I guess there is an error in your code. The reason for the change in the decision rules compared to first order can be found here [code]Bug in k_order_solver? ghx changes from order 1/2 to 3]