Prediction using a forvalues Loop

[Laura Rafferty]

I am trying to generate a value for the expected change in the unemployment rate (expdump) by creating a forvalues loop. There are a total of 892 periods in my data set. I would like each prediction to incorporate data from all of the previous periods, which is why I'm using a loop.

In this loop, I am trying to run the predict command and have it generate one new variable column in my data set called expdump, which would represent the expected change in unemployment. When I run the code below, though, it gives an error that says "expdump already defined."

Any advice on how to solve this issue?

Here is the code I have currently:

forvalues i= 180/891 {
regress dump ipgrate1 ipgrate2 ipgrate3 dump1 dtb3 dba if period in 1/`i'
predict expdump if period==`i'+1
}

Thank you!

[Nick Cox]

This kind of question is often addressed here e.g. the very recent thread https://www.statalist.org/forums/for...forvalues-loop

However, your twist

 if period in 1/`i'

is very confused about if and in. in refers only to observation numbers, not values of a variable. Also

if period

.

by itself is legal, but not what you want. It just restricts regression to non-zero values of period, which here is no restriction.

You may want something more like

gen expdump = .
forvalues i= 180/891 {
     tempvar work
     regress dump ipgrate1 ipgrate2 ipgrate3 dump1 dtb3 dba if inrange(period, 1, `i')
     predict `work'
     replace expdump = `work' if period == `i' + 1
     drop `work'
}

but without any kind of data example I am guessing.

[Laura Rafferty]

This works perfectly. Thank you so much!

[George Bateson]

Hi I was wondering how you could code forecast measures for this. E.g. How would you find the mean absolute error?

Thanks!