what is the test after ppml? help needed, thank you

[FEI ZHANG]

Dear Joao,
I am doing a Gravity Model estimating the export potential of China export to other countries. In my Gravity Model, there is one exporter (China), 85 importers, 17 years (N=1445).

My question:
(1) Is there problem of R2 (0.128), pplm ?
(2) should I use the result of ppml or "xereg ……, fe"(according to hausman re fe)? fit2 =0 test is rejected.
(3) when I use data of 28 importer countries, 17 years (N=476), R2=0.35, but with N=1445, R2=0.128, should I change regressors?
Thank you.
Fei

Details are as follows:

Equation:
Export=a+a1ln(popj)+a2ln(averagegdpj)+a3ln(yield)+ a4ln(distcap)+a5ln(exchangerate)+a6ln(comrelig)+a7 comlang_ethno+a8contig+a9fta_wto

Command:

• sum exp lgpop lgavgdp lgyield lgdist lgrelig lgrate comlang_ethno contig fta_wto
• ppml exp lgpop lgavgdp lgyield lgdist lgrelig lgrate comlang_ethno contig fta_wto

• predict fit, xb

• gen fit2=fit^2
• ppml exp lgpop lgavgdp lgyield lgdist lgrelig lgrate comlang_ethno contig fta_wto fit2
• test fit2=0

• eststo: xtreg lgy lgpop lgavgdp lgyield lgdist lgrelig lgrate comlang_ethno contig fta_wto,fe
• estimates store FE
• eststo: xtreg lgpop lgavgdp lgyield lgdist lgrelig lgrate comlang_ethno contig fta_wto ,re
• estimates store RE
• xttest0
• hausman FE RE,constant sigmamore

[Joao Santos Silva]

Apologies for the late reply.
1 - No, you can generally ignore the R2
2 - PPML is the preferred estimator, so use ppml or ppmlhdfe commands.
3 - The R2 is often higher in smaller samples; just use as much data as you can.

Best wishes,

Joao

[Qiyangfan Feng]

Dear Joao,

Is ppml a preferred modle to estmate effect on economic growth? There are some interaction terms in my equation.And the effect of them are not linear(is nonlinear).

Thanks,

Feng Qiyangfan

[Joao Santos Silva]

Dear Qiyangfan Feng,

I am not familiar with that literature; can you please give more details on the model you want to estimate?

Best wishes,

Joao

[Qiyangfan Feng]

Dear Joao,

The model I want to estimate is as follow:

Y=α+βX1+βX2+βX1*X2+βControl+δ。 As you know, X1*X2 is not a linearn term(But also not a "U" trend). Could ppml or ppmlhdfe be prefered to estimate the model?

Thanks!

Qiyangfan Feng

[Joao Santos Silva]

Dear Qiyangfan Feng,

That model is linear in the parameters, so do not use ppml or ppmlhdfe because these commands estimate exponential models. Of course, if in your mode Y is the log of something, then ppml and ppmlhdfe would be appropriate to estimate the model in levels.

Best wishes,

Joao

[Qiyangfan Feng]

Dear Joao,

I used the nuclear estimation to estimate the X2 marginal effect of X1 on Y , it shows that the effection is unlinear(Diminishing marginal return) . Should I treat the theoretical model as nolinear model,and do not list the specific econometric model. And then should I use ppml or ppmlhdfe model?

Thanks,

Qiyangfan Feng

[Joao Santos Silva]

Sorry, Qiyangfan Feng, I am not able to advise on that.

Best wishes,

Joao