weighted least squares?

[minson]

Hello,
I have run a linear regression in Eviews. I have done a cross-sectional regression. Every month or so, I add all the newest data to the workfile and rerun the regressions, just to have all the latest data included in the equations. But the model gives the same emphasis to data from the year 2006 that is does to the data from 2009.
Is there a way of making the data become less and less important as we go back in time? So that the curves fit the latest data more closely than it does the data from a long time ago.
Reading the weighted least squares help page has left me confused and unsure as to whether it's appropriate for what I want to do.
Any suggestions/ideas would be appreciated.

Thanks
Minson

[trubador]

Although weighted least squares (WLS) has several disadvantages, you can still use it. Of course, the most important issue here is to decide weight structure. As an example, you can do the following:

Code: Select all

series weight = @cnorm(3*(@trend-@mean(@trend))/@stdevp(@trend))

The function above will generate a S-shaped weight structure. You can adjust the shape (and therefore the weights) by changing the value of the multiplier. Other than that, you can try linearly increasing or binary weights. When you decide the weight structure, you can estimate WLS via supplying the weight series to "Estimate/Options/Weighted LS&TSLS" in the Equation window or directly from the command line:

Code: Select all

equation eq1.ls(w=weight) y c x1 x2