IVCLOGLOG v2.0.0: Updated Stata Command for Estimating a Complementary Log-Log Model with Endogenous Covariates

[William Liu]

Dear all,

I am happy to announce that ivcloglog has been updated to v2.0.0. Below is a list of all the changes:

1. Endogenous variables no longer have to be first-stage dependent variables.
   • The option endogenous() is now used to specify the names of the endogenous variables if any of them are not the same as the first-stage dependent variables and are instead functions of the first-stage dependent variables.
   • The original endogenous()option has been renamed to auxiliary(); use this to specify the first-stage (i.e., auxiliary) dependent variables.
2. Replay functionality: previous estimation output is replayed after inputting "ivcloglog".
3. Modified output for improved clarity.
4. Minimum version of Stata has been changed to be version 11 instead of version 14
5. Bug fixes and minor tweaks.

Here is an example of why allowing endogenous variables to be different from the first stage dependent variables can be very useful. Imagine you have data on loan default. A loan is underwater if the loan-to-value ratio (LTV) is greater than 100%:

underwater := 1(LTV>100%).

If underwater is an endogenous variable, the fact that it is a discrete variable could be problematic. Instrumenting a discrete variable in a nonlinear model necessarily requires a nonlinear first stage for validity, necessitating strong assumptions than using a linear first stage (Wooldridge, 2015). However, if LTV is an observed variable, then we can avoid instrumenting underwater. We can instrument LTV instead because the resulting control function terms will still be a valid control function for underwater by virtue of the fact that underwater is a function of LTV (Liu, 2023). Doing this allows for increased robustness by avoiding unnecessary assumptions.



Original thread: https://www.statalist.org/forums/for...ous-covariates
References:

• Liu, W. (2023) A theory guide: Instrumenting a discrete-data proportional hazards model with control functions. Working Paper. (Will be available on arXiv in the future.)
• Wooldridge, J. M. (2015). Control function methods in applied econometrics. Journal of Human Resources, 50(2), 420-445.