A quick look at the FITOPTIONS function gives the answer.
Both the linear and nonlinear least squares methods allow lower and upper bounds to be placed on your parameters.
help fitoptions
Were this NOT an option, then you could still solve the problem simply by squaring that parameter inside your model. Until the supreme court has ruled against this, the square of a number is always positive. They might do so though any day now, so beware. :)
