Overidentification test (Sargan–Hansen) for instrumental variables in a Dynamic panel model with RE

[Silvana Robone]

Hello,

I am estimating a dynamic panel model with random effects, and I am using the Mundlack correction to mimic fe. My main independent variable is likely to be endogenous and I am instrumenting it with two IV variables. I am using the command "xtivreg".
I cannot find the proper command in STATA to run an overidentification test (like Hansen-Sargan) when using the command "xtivreg". Any suggestion to fix this problem and run the overidentification test?
Many thanks in advance!
Cheers,

Silvana

[Manh Hoang Ba]

You can use the xtoverid command after running xtivreg to calculate the Sargan-Hansen statistic and p-value.
However, in dynamic models with individual effects (fixed or random), the lag of the dependent variable is also an endogenous variable, which introduces bias into the overall parameter estimates of the model.
With large T-panels, the bias can be ignored. With short T-panels, you need to instrument for the lag of the dependent variable, and internal instruments such as Blundell-Bond may be useful.

[Silvana Robone]

Dear Manh,

first of all, many thanks for your kind reply!
I have tried already to use the "xtoverid" command after having estimated my model, but I got the following error message:

xtoverid
o. operator not allowed
r(101);

maybe the command xtoverid doesn't work after "xtivreg". Or do you have other explanation?

As far as the issue of the lag of the dependent variable, What is a large T to you? I have 8 waves in my panel......

Thanks again for your advice :-)

Cheers,

Silvana

[Manh Hoang Ba]

xtoverid does not allow factor variables, so you need to create factor variables manually. For example, the factor variable is named fv_var, the following command will help you create dummy variables with names starting with fv:

tab fv_var, gen(fv)

With your data, it is much more convenient to use the xtabond2 or xtdpdgmm commands.