Questions on ARDL (Autoregressive Distributed Lag) regression

[Olivia Johns]

So, I have panel time series, and I ran ARDL regression on the data I have, I tried lag 3 just to see how the results come out.



. reg L(0/3).depvar L(0/3).indvar1 L(0/3).indvar2

Source | SS df MS Number of obs = 2,643
-------------+---------------------------------- F(11, 2631) = 54699.03
Model | 17287.6987 11 1571.60897 Prob > F = 0.0000
Residual | 75.5937191 2,631 .028731934 R-squared = 0.9956
-------------+---------------------------------- Adj R-squared = 0.9956
Total | 17363.2924 2,642 6.57202589 Root MSE = .1695

------------------------------------------------------------------------------
depvar | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
depvar |
L1. | 1.092625 .0202706 53.90 0.000 1.052877 1.132373
L2. | -.1279143 .0290344 -4.41 0.000 -.1848468 -.0709818
L3. | .0321368 .0201195 1.60 0.110 -.0073149 .0715885
|
indvar1|
--. | -.1302286 .0344263 -3.78 0.000 -.197734 -.0627232
L1. | .0136766 .0457565 0.30 0.765 -.0760458 .1033989
L2. | .0223419 .0457573 0.49 0.625 -.0673819 .1120658
L3. | -.0143141 .0327803 -0.44 0.662 -.0785919 .0499638
|
indvar2 |
--. | .2211192 .0331098 6.68 0.000 .1561953 .2860431
L1. | -.249598 .0459983 -5.43 0.000 -.3397945 -.1594014
L2. | -.0312808 .0429772 -0.73 0.467 -.1155534 .0529918
L3. | .0431742 .0283124 1.52 0.127 -.0123426 .0986909
|
_cons | .7006697 .1096082 6.39 0.000 .4857428 .9155967
------------------------------------------------------------------------------



==>This is the results I got, so based on the significant on each variable, I decided to run, the following




. reg L(0/2).depvar indvar1 L(0/1).indvar2

Source | SS df MS Number of obs = 2,797
-------------+---------------------------------- F(5, 2791) > 99999.00
Model | 18297.2294 5 3659.44589 Prob > F = 0.0000
Residual | 79.8522759 2,791 .028610633 R-squared = 0.9957
-------------+---------------------------------- Adj R-squared = 0.9956
Total | 18377.0817 2,796 6.57263295 Root MSE = .16915

------------------------------------------------------------------------------
depvar | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
depvar |
L1. | 1.085766 .0193572 56.09 0.000 1.04781 1.123722
L2. | -.0897081 .019362 -4.63 0.000 -.1276734 -.0517428
|
indvar1 | -.108408 .0168942 -6.42 0.000 -.1415345 -.0752816
|
indvar2 |
--. | .2370868 .032686 7.25 0.000 .1729957 .3011779
L1. | -.2527008 .0329392 -7.67 0.000 -.3172885 -.1881132
|
_cons | .6904025 .0996915 6.93 0.000 .4949261 .8858789
------------------------------------------------------------------------------

.
end of do-file

.

===> does this sound like a correct way to decide the lag length?

[Sebastian Kripfganz]

There is an earlier Statalist discussion about this topic:
Optimal lag length for ARDL in panel

Maybe it helps as a starting point.