Near Singular Matrix or Overflow - System Estimation

[whoami]

Hello.

I have a SYSTEM of 4 equations: Supply, Demand, Inventory and Price equation.
My objective is to estimate speculation (if any) in the pricing equation.
Note: The variable estimating speculative behavior is an explosive time series that is conditioned on other parameters of supply, demand and inventory.

1. OLS gives me sensible results when the variables pertaining to speculative behavior are NOT included. Most of the parameters are significant.

2. But I get weird results with OLS when the speculation variables are included
2a. sometimes Eviews spits out some results and none of the parameters are significant.
2b. sometimes I get the error OVERFLOW
2c. sometimes I get the error NEAR SINGULAR MATRIX

I think the numerical estimation is very unstable with the inclusion of additional variables that are explosive time series.

3. I get some results using FIML

When I do get estimates as in 1, 2a, and 3 -- the parametric estimates are almost identical, but the standard errors are very different. Do you have any suggestions with regard to proper estimation in this case?

Thanks
Subbu

[EViews Gareth]

Hard to say without seeing the data/specification.

[whoami]

Thanks for your quick reply.

Multiple runs of the system give me different parameter estimates with FIML. that is a correction of my earlier post above.