PPML with panel data

[Trang Truong]

Hi all,

I have a panel data an use gravity model for it, my data has zero trade flow, so I use PPML regression to estimate. Is this suitable with my data, or I must to use other methods?

Thank you for your help!

[Joao Santos Silva]

Hi there,

There is no problem at all with using PPML regression with panel data.

All the best,

Joao

[Arista Kusuma]

Hi Mr. Santos Silva,

May I know what statistical test (like heteroskedasticity, autocorrelation, etc) needed if we use PPML to estimate gravity model with panel data and disaggregated data?

Thank you.

[Guest]

Hello Arista

you may find this best practice paper for gravity estimation incredibly helpful that answers a lot of your questions (given you have the data needed):
Roberta Piermartini and Yoto V. Yotov 2016

https://www.wto.org/english/res_e/re...sd201610_e.pdf

Even though zero trade flows are a problem especially prevalent in disaggregated data you may also find a much better "punchline" why you favour the use of ppml in the paper above.

Kind Regards

[Joao Santos Silva]

Dear Arista,

PPML is quite robust. So, I just recommend that you use clustered standard errors and perform the RESET test.

Best regards,

Joao

[Arista Kusuma]

Dear Justus Meyer,

Thank you for your kind advice.

Best regards,

[Arista Kusuma]

Dear Mr. Santos Silva,

Thank you for your response. Do you it is necessary to use bootstrapped standard errors?

Best regards,

[Joao Santos Silva]

Dear Arista,

No, bootstrap has no obvious advantage in that case.

All the best,

Joao

[Arista Kusuma]

Dear Mr. Santos Silva,

Thank you very much!

Best regards,

[Arista Kusuma]

Dear Mr. Santos Silva,

In case of the gravity model with PPML estimator do not pass the Ramsey RESET model, what should we do?

My RESET test result:

( 1) [RealExport]fit2 = 0

chi2( 1) = 1.52
Prob > chi2 = 0.2180

my command:
ppml RealExport ln_G ln_S R ln_Distance ln_ExRate ln_Area Contig ComLangEt RCEP ASEAN, cluster(ID)
predict fit, xb
gen fit2=fit^2
ppml RealExport ln_G ln_S R ln_Distance ln_ExRate ln_Area Contig ComLangEt RCEP ASEAN fit2, cluster(ID)
test fit2=0


Thank you.

Best regards,

[Joao Santos Silva]

Dear Arista,

First, notice that what is being tested is the model, not the estimator. Second, the model passes the test because you do not reject the null of correct specification.

Best wishes,

Joao

[Tina Samuel]

I am also using a PPML model and I ran xtserial command to check for autocorrelation and it indicated that there's autocorrelation, is there's a way I could correct for this ?

[Joao Santos Silva]

Just use clustered standard errors.