ivppml and ivpoisson

[DN Jay]

Dear Stata community,


I have a panel gravity dataset. I have trying to run the following models

1. ivppml patents pc1_CS ln_gdp_o ln_gdp_d ln_wdi_gerp_qog_data_o ln_wdi_gerp_qog_data_d ln_tradeflow_baci Inst_Dist ln_dist contig comlang_off colony, inst( new_gendist_plurality ln_gdp_o ln_gdp_d ln_wdi_gerp_qog_data_o ln_wdi_gerp_qog_data_d ln_tradeflow_baci Inst_Dist ln_dist contig comlang_off colony)


what is the mistake I'm doing? when i try to include year_* origin_* and dest_* fixed effects the model does not converge. is this expected?

2. I have tried ivpoisson with the following code. However, i get an error saying "Hessian is not positive semidefinite". How can i correct for this?

ivpoisson gmm patents ln_gdp_o ln_gdp_d ln_wdi_gerp_qog_data_o ln_wdi_gerp_qog_data_d ln_tradeflow_baci Inst_Dist ln_dist contig comlang_off colony ( pc1_CS = new_gendist_plurality)

[Joao Santos Silva]

Dear DN Jay,

ivppml should converge, but check the discussion about convergence in the help file for ppml. Note, however, that these estimators are not valid when the model has fixed effects.

Best wishes,

Joao

[DN Jay]

Dear Joao Santos Silva ,

Thank you very much. I managed to get some acceptable results using IVPPML after clustering at the country pair level. when you say ivppml and ivpoisson do not provide valid estimates with fixed effects, is this because of a computational limitation within stata or its a model limitation?

[Joao Santos Silva]

It is because of the so-called incidental parameter problem.

[Michael Kwowove Didongo]

Good day, Prof. Please, I'm working on agricultural trade and trade integration in Africa. I want to use the negative iv-ppml estimator as a way to correct potential endogeneity but I don't have the command code to install. I would be glad if you could help me with the command code. Thank you

[Joao Santos Silva]

Dear Michael Kwowove Didongo,

I am not sure if I understand what you want, but maybe you are looking for this command; please check also the example here.

Best wishes,

Joao

[Michael Kwowove Didongo]

Thank you very much, Prof. I have used the IVPoisson estimation technique and my results are not statistically different from the PPML results. The results are good, I'm satisfied with them. Thank you.