Negative fitted values

#1
Hi,
I am trying to imput the prices of houses in a database using information on the carachteristics of the houses and a database in which i have the same characteristics and the prices of the houses.
When i run OLS i get negative fitted values both in my database as in the original database, but that doesnt make sense.

What can i do to avoid the negative fitted values?

Thank you in advance.

Laurie
 

Mean Joe

TS Contributor
#2
Is it impossible, or just doesn't make sense? Sometimes statistics reveals things which are counter-intuitive. Maybe some characteristic actually makes house prices fall...

If you must have positive fitted values, then you can reverse the value of that characteristic for all records. eg if the value is 4, then make it -4. Maybe the scale is just backwards.

If you don't like that, then you could just say that the value shouldn't be negative, so just set the fitted value to 0, ie drop that characteristic from the regression.
 

Dason

Ambassador to the humans
#3
Wait. Are we talking about fitted values as in predicted values? As in you plug in the x values for your original observations and it predicts a negative house price? Or are you talking about negative values for the coefficients in the model?