EViews.com

I am using a structural macro model for forecasting and am having trouble with the starting date for the forecast.
The typical case is that the actual data series have different sample period end points (the so-called ragged edge). I have a routine that uses the model to fill in these missing data points, so that all of the baseline (_0) variables in the model have the same end-point, say 2008 (the exogenous variables are extended though the entire forecast period). I do not touch the actual endogenous variables.
Then I solve the model starting in 2008 using:
Model.solveopt(i=p, e=f)
So starting values are set to the previous period's solution.

The problem is that I get a missing values error message in the 1st year of simulation (2008). When I check all of the variables used by the model - Proc/makegroup/all variables - there are no missing values before and including 2008.
If I start the simulation earlier it works fine.
Why does this happen?
And a guess. What starting values are used for lagged variables in the model. If its actuals, despite i=p, then that answers it .

Thanks for your help

Clive

It's often harder to understand a written description of these things than to see them in action, so it would be helpful if you could provide an example workfile with the model that is reporting the NAs during the solve. (Feel free to change the data values and/or names of variables).

One thing I can say is that missing starting values for the endongenous variables should not cause an NA error to be reported. If both the actuals and previous period's values are NAs, we'll set the starting values to something arbitrary (eg. 1.0) and try solving the model from there (which is not a bad startegy for linear or near linear models, but is unlikely to work for highly non-linear models).