rolling dfuller test for panel data - loop

[Luca Di Simone]

hi everyone,

My target is to perform a rolling dfuller test on panel data. Something like asreg or asrol for each panel and date of my data.

My sample is composed of N stocks and every stock has a specific value (from 1 to N) for "ID" variable and stock returns information "R" in different time/date ranges.

My sample has missing values and time numerical variable is "obst"

I am looking for a help about a specific nested loop. I would want to perfom, for every date (obst) and for every stock (ID), a dfuller test on last 250 stock returns (R) observations.
Then, I want to record some statistics (DFts, CV1,...) of the test stored in r().

This is my code:

gen DFts=.
gen CV1=.
gen CV5=.
gen CV10=.
gen DFpv=.

gen tmin=.
gen tmax=.

qui su ID

forvalues i = 1/`r(max)' {

qui su obst if ID == `i'
replace tmin=`r(min)'+250
replace tmax=`r(max)'

forvalues t = tmin/tmax {

qui dfuller R if ID == `i' in -249/`t'

replace DFts=r(Zt) if ID==`i' & obst==`t'
replace CV1=r(cv_1) if ID==`i' & obst==`t'
replace CV5=r(cv_5) if ID==`i' & obst==`t'
replace CV10=r(cv_10) if ID==`i' & obst==`t'
replace DFpv=r(p) if ID==`i' & obst==`t'

}
}

this code yields an error message

"invalid syntax
r(198);"



I hope someone can help me.

thanks

[Nick Cox]

Repeated significance testing on overlapping chunks of time series?

That's not what Stata is objecting to. I think Stata needs here something more like

local tmin=`r(min)'+250

forval t = `tmin'/`r(max)' {

The variables tmin and tmax are not needed at all.

In short, forval needs to be given scalars, not variables.

[Luca Di Simone]

Nick, thanks for your answer.

"Repeated significance testing on overlapping chunks of time series?"

yes, it is exactly in this view.

With your changes, now it's ok.

[Nick Cox]

Glad that the syntax works. Goodness knows what to make of numerous highly dependent significance tests.