prediction, unexpected results

[John Baxter]

Hi,

I run a prediction by the first 20 sample (y and x), and I would like to use the results to predict the 25th y_f (a special out of sample prediction). I think this is what I do in the programming, but it gives me the prediction for the 20th sample (in-sample). Why? How can I modify my program to obtain the out-of-sample (25th) prediction?

I have a loop, and I expect that it will give me prediction for 25th, 26th, 27th.

clear all

import excel "xyz1.xlsx", firstrow case(lower) clear

sca start_T = 20
sca n1 = 5
egen t1 = seq()
gen y_f = .

forvalues i = `=start_T'/`=start_T+2'{
reg y xa if t1<= `i' & t1> `i'-start_T // it works
predict temp in `i+n1'
replace y_f = temp in `i+n1'
drop temp
}


Thank you very much!!

[Nick Cox]

As before, you are asking what turn out to be similar questions in different threads without any cross-referencing.

Again, as explained one way or the other in earlier threads.

`i+n1'

is fantasy syntax. Stata won't add scalars on the fly like that. See e,g. https://www.statalist.org/forums/for...ubsample-by-in


To debug, you need to work with simpler examples

. local i 20

. scalar n1 = 5

. di `i+n1'
20

I agree with anyone thinking that Stata's reaction here is puzzling, even unhelpful. It just ignores the +n1 in the macro reference as so much trailing junk.

But, to back up, the start of the syntax here is that macname within a local macro reference should be a local macro name.

`macname'

What you then learn is that is not a bug to refer to a local macro that doesn't exist: Stata usually just evaluates it as an empty string. (Whether the result of that is legal and correct is a different story).

What is biting here is that your fantasy syntax isn't construed as illegal but it has a weird consequence, which can only be understood by experiment, or so I suggest. Whether you've unearthed a bug in Stata that -- to be as positive as possible -- won't bite often is a different question.

[George Ford]

local i 20
scalar n1 = 5
di `i+n1'
20
di `i' + n1
25