Using regressions in monte carlo simulations

Hey Guys,

I'm building a valuation model with monte carlo and regression elements but could use a bit of help.

The model breaks down the firm's revenue and costs by product. To estimate the cost of producing each product (cost per unit), I've regressed unit sales of each product against the firm's total direct costs. The coefficients look good, but I have three questions.

First, a basic one. For each coefficient's probability distribution, can I use the standard error or do I need some other measure of deviation? Intuitively, the standard error seems right to me. But, the book I'm reading doesn't address this problem directly (I'm trying to teach myself this stuff, and it's a bit tough.), and I want to make sure.

The second question is a bit more complicated. Because I'm regressing the dollar value of costs against raw unit sales, my sample has a problem with heteroscedasticity. (At least, I'm pretty sure it does. I'm still wrapping my head around the various diagnostic tests for this.) I know I could use the logs of the variables, but dollar values are better for explanatory purposes. Would it be valid to use the logs in a separate regression to estimate the standard error as a percent of the coefficient and then apply this percent to the original regression to correct for the heteroscedasticity?

Finally, I have a question about the constant. According to my book, it's rarely a good idea to set the constant to zero because that could create a severe bias. However, in this case, I know it should be zero because if the firm doesn't sell any units, it won't have to pay any production costs. (Operating costs, on the other hand, are a bit more complicated because many of the costs are fixed.) Do you think it's acceptable to exclude the constant in this scenario?

Like I said, I'm still learning, so if I've missed something basic or misused any of the terms, please let me know.

Thanks in advance