I am currently performing MS-VAR on givewin2, but many models cannot be partitioned when doing impulse response, and I also want to perform partitioning Granger causality and variance decomposition. Could you please give me some guidance or provide some code
Well, you know, I would argue that you can obtain conditional IRFs ie, conditional to a particular state inside the state space, by using the model equations. This idea asumes you will stay in such a state during the whole (forecasting) horizon. But clearly, this is a hard assumption, so people have propose alternative sophisticated methods. From the economics perspective, initial proposals were:
.Potter, Simon (2000) Nonlinear Impulse Response Functions, Journal of Economic Dynamics and Control Vol. 24, No. 10, pgs. 1425-1446.
.Weise, Charles (1999) The Asymmetric Effects of Monetary Policy: A Nonlinear Vector Autoregression Approach, Journal of Money, Credit and Banking, Vol. 31, No. 1, pgs. 85-108.