historical decomposition in VAR with time-varying parameters

[tclark]

I need to code up an historical decomposition for a VAR with time-varying parameters. Specifically, I need to decompose the variation over time in the actual data into the contributions from the structural shocks of the model. The difficult (for me, anyway) aspect of the problem is the time-variation in the parameters. One approach would be to loop over time, assigning the coefficients of the VAR at each point in time t, and using the FORECAST command to compute one-step ahead forecasts for each t on the basis of the particular shock at t, and then store the forecast for period t as the actual value in t (I would need to do this for each shock of interest). Is there some other approach that would be more efficient? Any suggestions would be appreciated.