How to concisely generate many lagged variables

[Alex Williamson]

Hello all,

I am trying to generate a large number of lagged variables, all lagged once, and all mostly the same.

I have 24 variables which exist, all of which I am trying to lag into new variables - only lagging each by one level. The code I am using for each individually is

bys country: gen l_[variable] = l.[variable]

Do I need to write this out 24 times or is it possible to write a concise version of this? Thanks!

[Andrew Musau]

xtset country year
foreach var in var_a var_b var_c var_d var_e{
    gen l_`var'= L.`var'
}

In most cases, this is unnecessary as Stata supports time-series operators for most estimation commands. For example, if I need to regress the lags of mvalue and kstock on invest in the Grunfeld dataset:

webuse grunfeld, clear
xtset company year
regress invest L.(mvalue kstock)

Res.:

. regress invest L.(mvalue kstock)

      Source |       SS           df       MS      Number of obs   =       190
-------------+----------------------------------   F(2, 187)       =    340.11
       Model |  7220371.82         2  3610185.91   Prob > F        =    0.0000
    Residual |  1984955.18       187  10614.7336   R-squared       =    0.7844
-------------+----------------------------------   Adj R-squared   =    0.7821
       Total |     9205327       189  48705.4339   Root MSE        =    103.03

------------------------------------------------------------------------------
      invest |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |
         L1. |   .1212423   .0064621    18.76   0.000     .1084943    .1339903
             |
      kstock |
         L1. |   .2503195   .0316839     7.90   0.000     .1878156    .3128233
             |
       _cons |  -43.29118   10.93372    -3.96   0.000    -64.86048   -21.72188
------------------------------------------------------------------------------

See

help tsvarlist

[Alex Williamson]

Thanks!
My one question is whether this would be possible with panel data? I forgot to mention, but the reason that it is -bys country- is that this is a panel dataset sorted by year by country - hence I was generating new variables rather than doing it within the regression itself as I was not sure how to get Stata to account for the groups.

[Andrew Musau]

The xtset command already declares that the data is a panel to Stata. So Stata does take into account the panel structure of the data when creating lags. See

help xtset

[Alex Williamson]

Ah yes, my mistake! Thank you so much for your help!