Errors specification in a state-space model.

[argen]

I'm trying to estimate the parameters of the flexible accelerator model of induced investment on regional level.
The usual flexible accelerator model looks like

INV_IND = c(1)*Y + c(2)*Y(-1) + c(3)*INV_IND(-1) + c(4)*INV_IND(-2)

where INV_IND is induced investment, and Y is observable output (adaptive output expectations are assumed).
The ARDL model with data on regional gross investment and output gives poor estimates because autonomous investment is also included in gross investment.
Autonomous investments are almost impossible to model at the regional level, since they are usually determined by extra-regional factors.(They can be seen as occasional investments for the region)
I have an idea to use the state-space model to get estimates of the parameters of the regional flexible accelerator model.
The idea is pretty simple. The signal equation defines gross investment as the sum of induced and autonomous investment.
Both types of investments are unobservable, but we have a recursive model according to which the induced part of gross investment are made.
This model, which is given above, will be represented in the state equation (two equations).
Then the residual term in the signal equation must represent an autonomous investment. But residuals in the model must all be non-negative, with some positive expected value.
How can I specify the error term in the signal equation so that it has the properties I need? Is it possible?
If not, could another model specification be possible?
The preliminary version of the model is as follows:

@signal INV_GROSS = c(1) + c(2)*INV_IND + [error term = INV_AUT]
@state INV_IND = c(3)*Y + c(4)*Y(-1) +c(5)*INV_IND(-1) +c(6)*INV_IND_LAG(-1) + [var = exp(c(7))]
@state INV_IND_LAG = INV_IND(-1)


c(1)=0; c(2)=1

Perhaps autonomous investments should also be modeled in the state equation, but I can not guess how to do this.

[EViews Glenn]

That error restriction isn't possible.