Dear Statalist,
I’m writing my bachelor thesis at the moment. I want to reproduce a paper about the Phillips Curve relationship in the forecasts of professional forecasters. But I don´t know which estimation method I should use. I already informed me a lot but didn´t find what I am looking for.
Unfortunately, I haven´t the do-File and I didn´t get response to my emails send to the author. So I decided to ask here for advice.
My data are panel data and have long-format with N>T and are unbalanced. I include a year dummy for every single year to control for time fixed-effects.
In the paper there is written: “we applied the fixed-effect estimator when the Hausman test rejects the null hypothesis, otherwise the random effect estimator was applied”. On plus cause of overlapping forecast horizons we have serial correlation in the error term by construction. To overcome this the author uses the serial correlation model AR(1) to account for the autocorrelation in the error term. Additionally, cause of heteroskedasticity we have to use robust standard errors.
So I am looking for an estimation method which I can apply for random and fixed effects and where I can also include AR(1) and use robust standard errors to control for heteroskedasticity.
Anyone konws a method like this? Or can I estimate fist without Robust Standard Errors and include them afterwards - is there a possibility?
Here is a short summary, what I have already tried:
This is my first post in this forum, so don´t hesitate to let me know if there is missing some information for a proper answer.
Best regards and thanks in advance!
Hannah
I’m writing my bachelor thesis at the moment. I want to reproduce a paper about the Phillips Curve relationship in the forecasts of professional forecasters. But I don´t know which estimation method I should use. I already informed me a lot but didn´t find what I am looking for.
Unfortunately, I haven´t the do-File and I didn´t get response to my emails send to the author. So I decided to ask here for advice.
My data are panel data and have long-format with N>T and are unbalanced. I include a year dummy for every single year to control for time fixed-effects.
In the paper there is written: “we applied the fixed-effect estimator when the Hausman test rejects the null hypothesis, otherwise the random effect estimator was applied”. On plus cause of overlapping forecast horizons we have serial correlation in the error term by construction. To overcome this the author uses the serial correlation model AR(1) to account for the autocorrelation in the error term. Additionally, cause of heteroskedasticity we have to use robust standard errors.
So I am looking for an estimation method which I can apply for random and fixed effects and where I can also include AR(1) and use robust standard errors to control for heteroskedasticity.
Anyone konws a method like this? Or can I estimate fist without Robust Standard Errors and include them afterwards - is there a possibility?
Here is a short summary, what I have already tried:
- Xtgls: no option cause of N>T
- Xtreg: don´t allows for ar(1)
- Xtregar: don´t allows for robust standard errors
- ...
This is my first post in this forum, so don´t hesitate to let me know if there is missing some information for a proper answer.
Best regards and thanks in advance!
Hannah
Comment