not sorted error when combining bysort with xtreg

[Dan Daugaard]

I am getting a "not sorted error" when attempting to run a time series regression for different "styles" (ie different categories) of funds. Any suggestions on applying the bysort would be appreciated - or suggestions on alternative approaches to conducting the exercise would also be appreciated. Thank you, Dan
The data looks like

* Example generated by -dataex-. To install: ssc install dataex
clear
input long id float month str1 retail_fund str4 style float(age ff size)
90727 696 "Y" "EDYG" 10      .47116  -.5108256
90727 697 "Y" "EDYG" 11   -.1180397  -.6931472
90727 698 "Y" "EDYG" 12      .41804   -.356675
90727 699 "Y" "EDYG" 13   .13305917 -.22314353
90727 700 "Y" "EDYG" 14    -.023651 -.22314353
90727 701 "Y" "EDYG" 15    -.001185 -.22314353
90913 696 "Y" "EDYG" 10   .12215516  2.6602595
90913 697 "Y" "EDYG" 11 .0046509327   2.631889
90913 698 "Y" "EDYG" 12     .103537   2.714695
90913 699 "Y" "EDYG" 13   .07039671   2.785011
90913 700 "Y" "EDYG" 14  .015616804   2.827314
90913 701 "Y" "EDYG" 15   .03278398   2.862201
end
format %tmYYMon month

and the code for running the regressions across styles is

bysort style: xtreg ff L.ff size age if retail_fund=="Y", re

(there are other funds in the full data set - ie some which are not retail and other types of style - but I hope this sample of the data helps set out the idea) thanks again, Dan

[Clyde Schechter]

You don't show your -xtset- command. I assume it was -xtset id month-.

Your problem is that in order to use the lag operator L., the data must be sorted by the same variables that defined the -xtset-. Because you sorted by style for this command, you scrambled the data and cannot use the lag operator.

There are a few ways to work around this. The most direct is to simply create a lagged ff variable and then abandon the use of the lag operator in the -xtreg- command:

gen lff = L.ff
bysort style: xtreg lff size age if retail_fund == "Y", re

Another approach is to do the regressions inside a loop over style

levelsof style, local(styles)
xtset
foreach s of local styles {
    xtreg L.ff size age if retail_fund == "Y" & style == `"`s'"'
}

Note that these two approaches will produce the same, correct, results if each style-based group of observations contains the full range of observations for each id that appears in the style based group. (That is, style must always be constant within each group defined by a value of id.) If that assumption is not true, then the two approaches will give different results because their estimation samples will differ. And you would have to figure out which approach is what you need.

[Dan Daugaard]

Thank you very much Clyde, yes, sorry, I forgot to include the -xtset- command and, yes, you are correct it was -xtset id month-
Thank you for your solutions - very effective, thank you also for the insight into what was going wrong.
One tiny dilemma I've hit in wrapping the exercise up,
I can save the results to a new file, eg

save demo_data1.dta

but when I update the data, eg

replace demo_data1.dta

I get a "factor-variable and time-series operators not allowed" error?
Sorry if it's something obvious, thank you, Dan

[Steve Samuels]

replace is for replacing variable values. The correct command is:

save demo_data1, replace

[Dan Daugaard]

yes, of course it is, thank you very much Steve, I just had a minor brain freeze, sorry, Dan