Hi, If you run a regression like Y= XB + GH + error where GH is unobserved so the actual regression you will run is Y = XB + (GH + error) and then use the output to forecast various things. If G has variables in it that trend in one direction or another over time ( i.e population in a given area) then will the forecast get less accurate as we get away from the time period used to estimate the coefficients and if so, is there some test that can be used to give an idea of how often the model should be re-estimated? This would also mean that the R^2 should decrease as we get away from the time period used for estimation right? Thanks, Chris