Significance of variables in VAR model and forecasting

[Bablowski]

Hello,

I would like to ask how to solve presence of statistically insignificant variables in VAR model.
I constructed VAR(4) model for two variables. Diagnostic control of residuals was satisfying. Nevertheless, there are some statistically insignificant variables in the models. I want to make forecast from this system of models. When I remove these insignificant variables, the forecasts are very poor and Root Mean Square Error very high compared to values of response variable. Should I forecast only from model with all variables included?
I think that presence of statistically insignificant variables in model does not necessarily mean that the model is not good. In standard one equation regression model I would throw away insignificant variable but in VAR model it is not so easy as for order 4 I cannot throw away variables of lag 3 when lag 4 is significant.

Thank you for any comments.

[startz]

Keep the variables. And throwing away insignificant variables in a regression is wrong.