Statsby & Loop

[alegiov]

Dear members,

I'm trying to combine the command "statsby" in a loop to obtain a new dataset composed only by the R2 derived from equations done on this dataset

This is how my dataset looks like:
YEAR | VAR A | VAR B | VAR C
1 | z | y | d
2 | z | y | d
3 | z | y | d
4 | z | y | d
5 | z | y | d
6 | z | y | d
For each year and each VAR, I would like to run the following AR(2) equation: reg VARx L1.VARx L2.VARx and store the R2 of each equation to derive a new dataset that would look like:

YEAR | VAR A | VAR B | VAR C
1 | | |
2 | | |
3 | r2(y3a) | r2(y3b) | r2(y3c)
4 | r2(y4a) | r2(y4b) | r2(y4c)
5 | r2(y5a) | r2(y5b) | r2(y5c)
6 | r2(y6a) | r2(y6b) | r2(y6c)

I've therefore created created the following loop:

________________________

foreach var of varlist _all {
local vl = "`vl' `var'"
}
local exclude "year"
local vl: list vl - exclude // this is to remove year from the var list

foreach var of local vl {
preserve
keep year `var' // this is to consider only one variable per time
tsset year


levelsof year, local(levels)
foreach x of local levels {
statsby e(r2), by(year) clear: `var' L1.`var' L2.`var' if year== `x' // this should create the R2 for each year
}
restore
}

________________________


unfortunately it doesn't work. Could you please give me some advice?

many thanks,

Alessandro Giovannini

[Joe Canner]

Alessandro,

There may be other issues, but right away I can see that your statsby command is missing the name of the regression command you want to run. Try:

statsby e(r2), by(year) clear: reg `var' L1.`var' L2.`var' if year== `x'

If this doesn't help, then you need to be more specific when you say "it doesn't work".

Regards,
Joe

[Roberto Ferrer]

What does "doesn't work" mean? Post exact output to increase your chances of getting help.

Guesses are:

You did not tsset your data
You are getting an insufficient observations error due to the if qualifier (if year == `x')

What does it mean to run a regression for a particular year using time series data? It seems like you're trying to do some kind of rolling regression using only three data points: x_t x_t-1 x_t-2.
Other details could probably be improved (e.g. the use of preserve / restore).

See

help tsset
help corrgram
help rolling

and more generally, help time series.

Consider also help postfile.

[Roberto Ferrer]

I can see that your statsby command is missing the name of the regression command you want to run.

Very important point.