Generating Predicted Values By performing separate regressions by year

[Chinmay Sharma]

Hi All,

I have a dataset that resembles the following:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(year GDP employment)
1990  12 123
1990 321  23
1990 123 213
1990 123  32
1990 123 213
1991 123 123
1991   3   2
1991  21  23
1991 213 321
1991  21 123
end

In the above, data on GDP and employment is stored by year. There is no reason to believe that the relationship between GDP and employment is stable over the course of the various years in my dataset. As such, I wish to estimate a separate regression of GDP on employment (regress GDP employment) by year, and generate predicted values for GDP based on the regression to be stored in the same column, call it Predicted. As such, the final dataset should have an extra column wherein each cell corresponds to the fitted value of GDP based on the value of employment obtained from the beta specific to the regression in that year. Are there any ways to do this besides resorting to mata?

Many thanks.

[Chinmay Sharma]

I think its quite straightforward: I did the following:

gen fitted=.

levelsof year, local(levels)
foreach i of local levels {
regress GDP employment
predict temp
replace fitted=temp if year == `i'
drop temp
}


I think this works.