Problem with Panel ARDL MG PMG

[salma bibi]

i am working on oil price shocks where i have divided price shocks into positive and negative shocks .i have taken sample of oil importing economies divided into 3 groups on the basis of oil imports A(0-10%oil imports)=22 B(10-20%)= 45 C(20-30%)=12 .the dependent variable is industrial production(lip) independent variables in the analysis are exchange rate(lex) (both used in log form), oil price shocks negative (opn),oil price shocks positive (opp)
i am using panel ARDL technique using syntax(stata 14)
xtpmg d.lip d.opn d.opp d.lex, ec(ec) lr(l.lip opn opp lex ) mg replace
expression (-_b[opn]/_b[L.lip]) evaluates to missing
r(498);
the above error appears for group A and B and not for C group.
suggestion required .

[JanDitzen]

The error implies that either _b[opn] or _b[L.lip] is missing. That is the coefficient of the corresponding variables cannot be estimated. Reasons can be multicollinearity or missing values in the data. My suggestion would be to run the model in the two stages, i.e. once with pooled coefficients (simple OLS with the variables from the lr() option) and secondly the short run coefficients via a mean group estimation.

[salma bibi]

highly obliged for your precious time multicollinearity might be reason (as there is no missing value for opn and opp) as in syntax where i drop negative shocks(opn) or positive shocks(opp) from the equation i get results
however your suggestion about estimation in two stages is not clear to me what syntax i will use for first part and second part any stata journal guidance please will be highly appreciated

[JanDitzen]

I was thinking of something like:

reg l.lip opn opp lex
xtpmg d.lip d.opn d.opp d.lex, mg

[salma bibi]

the first regression gives output by second code i get answer option ec() required
r(198);
as i insert ec at it shows the below given
xtpmg d.lip d.lop d.lex, ec(ec) mg replace
invalid syntax
r(198);
will be highly obliged if my query is solved .

[Hamid muili]

Dear Prof JanDitzen , I will like to know if there's an assumption concerning the numbers of Cross sectional unit and time dimension for panel ARDL. I have seen it used with different specification both for N<T and N>T. Based on the premise of large N and Large T.

[JanDitzen]

I answered your question in the cross-post here: https://www.statalist.org/forums/for...-of-panel-ardl