Weird Overflow/Log nonpositive Bug/Query

[CharlieEVIEWS]

Dear all,

I am getting an error from one part of a large computational routine which estimates and forecasts from an extremely large number of models. However, one object is confusing me: when forecasting from the VAR object (VAR forecast add-in) in the attached (separated from the main workfile) I get one error message (overflow) for forecasting from the model when the specification has one lag, and another error message (log of a non-positive number) when estimating larger lags. I'm clear on the fact that the overflow relates to extremely large numbers (explosive systems), but why does the error message change when the lag length of the specification changes? In addition to this, while my knowledge of numerical analysis is patchy at best, aren't we just doing a determinstic solve here, creating a one-step forecast from the coefficients, and then using these one step forecasts to make the second step, up to h-steps (i.e. an 'iterated' as opposed to 'direct' forecast)? If any member of the EViews community or seasoned econometrician would be so willing to enhance my understanding of this problem, I would be most eternally grateful,

Best wishes,

Charlie