[Joao Santos Silva]

Dear Joana Passinhas,

You can try to address the convergence problems by changing the relevant options; please check the help file. It may also be that the model you are estimating it too complex for the data you have. You are not including fixed effects, are you?

Best wishes,

Joao

[Joana Passinhas]

Dear Joao Santos Silva,

Thank you for your answer.

I am not including fixed effects.

I have tried to estimate the model with less observations (just cleared the dataset from n_>5000) and when I do that it does not produce the same warning. Could it be that I am using a too large of a panel?

Best,
Joana

[Joao Santos Silva]

Dear Joana Passinhas,

Too large a panel should not be an issue. Did you try changing optimization the options?

Best wishes,

Joao

[Joana Passinhas]

Dear Joao Santos Silva,

I have not yet tried changing the optimization since it takes around 3 days to run the estimation. Do you have any tip on what I could try to change first in the options?

Best,
Joana

[Joao Santos Silva]

Dear Joana Passinhas,

Given that you get convergence with a smaller sample, I would use those results as starting values. Otherwise, try the option mu. Feel free to contact me by email do discuss this in more detail.

Best wishes,

Joao

[Rajarshi Chowdhury]

Dear professor Joao,

I am using quantile regression estimation in stata instrumenting an endogenous variable along with a list of covariates. As I am using Annual Survey of Industries database, I have firm level data over a period of 10 years. But the dataset is not panel. Its a pooled cross section database.I want to know whether ivqreg2 command will work with pooled cross-section data? I have ran the command with full specification, but its is loading, not showing any result..

Best Regards,

Rajarshi

[Joao Santos Silva]

Dear Rajarshi Chowdhury,

The ivqreg2 command is suitable for the kind of data you have (it is not suitable for panel data models with fixed effects), but estimation may take a long time if you have a lot of data. As I suggested above, maybe you can obtain results with a sub-sample and use those as starting values for the full sample estimation.

Best wishes,

Joao

[Rajarshi Chowdhury]

Dear professor Joao,

I have run ivqreg2 to get the estimates of the coefficient of the explanatory variables in my model. now is there any way to graph the estimates at different quantiles obtained by using ivqreg2?
In case of ordinary quantile regression using "qreg" we can plot the estimates using "grqreg"..Can we do the same for ivqreg2?

[Joao Santos Silva]

Please try qregplot.

[Zubai Khan]

Joao Santos Silva Just a simple question: I need to interpret the test statistics of ivqreg2 but could not find anything on the web or paper. Can you help me with that? Particularly: GMM criterion Q(b), J-test. df, p-value. Moreover, which of these are essential to be reported? Also issues in outreg2 of the statistics.

[Joao Santos Silva]

Dear Zubai Khan,

The GMM criterion is the basis of the J-test for overidentifying restrictions. Notice that the estimator is based on a GMM estimator, so that is the usual J-Test. It is not essential to report it, but it is useful

Best wishes,

Joao

[Zubai Khan]

Joao Santos Silva Sorry to bother but I am sharing my experience while finding that ivqreg2 is sensitive to the ordering of exogenous variables & instruments and to get confirmation from you if I am correct. ivqreg2 will not generate a valid response (and will take loads of time) if you change the order of the control/exogenous variables while putting them in instruments. For instance, the following will be the correct model:

ivqreg2 y x1 x2 x3 x4 inst(z x2 x3 x4) q(.5)

or

ivqreg2 y x2 x3 x4 x1 inst(z x2 x3 x4) q(.5)

where x1 is endogenous. The change like the below will be incorrect and will create a problem:

ivqreg2 y x1 x3 x2 x4 inst(z x2 x3 x4) q(.5)

. It gave me hard time as I am used to xtabond2 which is not sensitive to the ordering of exogenous control variables instruments.

[Joao Santos Silva]

Dear Zubai Khan,

Although I am not sure of it, it may well be the case that you have to use the variables in the order you have in the first line, that is explanatory variables are in the same order, with the instrument replacing the endogenous explanatory variable. I'll try to look into this next time I update the command.

Best wishes and thanks for this,

Joao

[Joao Santos Silva]

Dear Zubai Khan,

I looked a bit more into this and the problem appears to be with the choice of default starting values. By default this is done with the old ivreg command, which appears to have a strange behaviour. Please check if your results are sensitive to the ordering of the variables when you use the mu option.

Anyway, I now have an updated version and would be grateful if you could test it. Please contact me by email if you are willing to do it.

Best wishes and thanks for pointing this out.

Joao

[Joao Santos Silva]

Dear All,

With big thanks to Kit Baum, an updated version of ivqreg2 is now available on SSC. This version fixes the bug that Zubai Khan identified and all users are encouraged to update.

Please do let me know if you find any problems with the new version.

Best wishes,

Joao