Instrumental Variables for GDP per capita

[Ben Pickering]

Hi there,

I'm currently doing my Economics undergraduate dissertation and have run into some trouble finding instrumental variables for my cross-country panel data model. My model's Y variable is education and X variable is Inequality, with interaction terms included for Institutional Quality, Country Development Level (essentially GDP per capita), and borrowing constraints (faced in a country). Intuitively, all of my X's are endogenous. I'm thus looking for instrumental variables (in panel data form not cross-section form) for:

• Inequality
• Institutional Quality
• Country development level (GDP per capita)
• Borrowing constraints

I've looked through the literature and can't find any past IVs used for these variables (in panel form), which is strange considered their popularity.

If anyone has any ideas or can point me to any papers in the literature which have used such IVs, I would be hugely grateful!

Many thanks,

Ben

[River Huang]

Hi Ben, To my experience, it is very DIFFICULT to find suitable IVs when we are considering panel data. That is probably the reason (whether valid or not) that people rely upon the dynamic panel data model using system/difference GMM (which uses internal instruments), Please check out (ssc install) xtabond2 command.

[River Huang]

Hi, Ben, Indeed, I also somehow suspect whether GMM is suitable in these cases, despite that it is popular in the literature (I believe the reason is because it is difficult to find valid IVs). As such, you might to check out the approach by Lewbel (2012, JBES), which relies upon the heteroskedastic errors to obtain identification. (“Using Heteroscedasticity to Identify and Estimate Mismeasured and Endogenous Regressor Models,” by Arthur Lewbel, Journal of Business and Economic Statistics 2012, 30, 67-80.) You might also to check out the -ivreg2h- command for implementing the estimator.