Recently saw this tweet from Jeffrey Wooldridge and it made me question my understanding of how spxtregress,resarpanel models are capable of. Wooldridge basically said that Stata's in-house spatial model package is completely incapable of properly fitting a regression model to data that are plagued by serial correlation.
Does the sarpanel estimator for Stata's spxtregress command not account for time-wise autocorrelation at all?
If not, is there anything we can use instead of the spxtregress command or in conjunction with the spxtregress command?
My understanding was that this particular specification of Stata's in-house spatial panel data models was capable of handling serial correlation within panels because the Stata documentation states the "sarpanel estimator was originally developed by Kapoor, Kelejian, and Prucha (2007)." Also, Kapoor, Kelejian, and Prucha stated in that cited paper that the assumptions underlying their estimator “imply that the model disturbances are…both spatially and time-wise autocorrelated.”
I'm aware of xsmle but is there anything I can use that has more flexibility than xsmle (e.g., I need to estimate a SAC model with random effects, xsmle can't do that) or can be used in conjunction with spxtregress?
I'm fairly invested in how much time I've spent diving into how spxtregress works and would like to be able to still use it if possible. I don't really have time to learn a whole new suite of commands and I need to get this study done.
Does the sarpanel estimator for Stata's spxtregress command not account for time-wise autocorrelation at all?
If not, is there anything we can use instead of the spxtregress command or in conjunction with the spxtregress command?
My understanding was that this particular specification of Stata's in-house spatial panel data models was capable of handling serial correlation within panels because the Stata documentation states the "sarpanel estimator was originally developed by Kapoor, Kelejian, and Prucha (2007)." Also, Kapoor, Kelejian, and Prucha stated in that cited paper that the assumptions underlying their estimator “imply that the model disturbances are…both spatially and time-wise autocorrelated.”
I'm aware of xsmle but is there anything I can use that has more flexibility than xsmle (e.g., I need to estimate a SAC model with random effects, xsmle can't do that) or can be used in conjunction with spxtregress?
I'm fairly invested in how much time I've spent diving into how spxtregress works and would like to be able to still use it if possible. I don't really have time to learn a whole new suite of commands and I need to get this study done.
