[Jesse Wursten]

Here is the updated xtistest

Best way to call it is by

xtistest, lags(2)

[Andreas Orp]

Ok, now I am able to do some kind of test, but I don't know if this is sufficient because the lags may be between several months and not the same lag for all observations. If i only do the FE-estimation with cluster, will this be a good way to control for potential autocorrelation? Also, as I have returns as dependent variable, I i believe that I have done what is normal to do for time series of stock prices to get a stationary process. What I want to find is whether trading based on different factors based on stock price and company specific factors will be statistical significant on the returns and may give excess returns.

[Marcus Fiedler]

Hello all,

1) I have a problem with xtset and I hope you can help me. I have to run a negative binomial regression because I have count data which has an exponential distribution.
When I was trying "xtset ID Year" (year range from 2005 to 2013), STATA called me that I have gaps but i don´t have any gaps in my data. Do you know what happened?
I have 93 IDs and 709 Years, unbalanced panel.

2) When I am trying xtset ID Year and I will include time dummies, is it possible or complicated? Or would it be better with xtset ID and then include time dummies?

3) I have 20 sector dummies and that could be a problem for my regressions. Up to now I have two levels (93 company and 709 observations) and 20 sector dummies. Could be a solution that I include a third level?
1 sector-ID
2 company-ID (93 IDs)
3 observations

Thank you and best regards,
Marcus

[Andreas Orp]

Thank you for the test Jesse! I will try with this as well.

[Jesse Wursten]

Ok, now I am able to do some kind of test, but I don't know if this is sufficient because the lags may be between several months and not the same lag for all observations. If i only do the FE-estimation with cluster, will this be a good way to control for potential autocorrelation? Also, as I have returns as dependent variable, I i believe that I have done what is normal to do for time series of stock prices to get a stationary process. What I want to find is whether trading based on different factors based on stock price and company specific factors will be statistical significant on the returns and may give excess returns.

I suggest looking at the literature. If everyone just uses a static fixed effects regressions when trying to predict returns, you're good to go.

[Jesse Wursten]

Hello all,

1) I have a problem with xtset and I hope you can help me. I have to run a negative binomial regression because I have count data which has an exponential distribution.
When I was trying "xtset ID Year" (year range from 2005 to 2013), STATA called me that I have gaps but i don´t have any gaps in my data. Do you know what happened?
I have 93 IDs and 709 Years, unbalanced panel.

2) When I am trying xtset ID Year and I will include time dummies, is it possible or complicated? Or would it be better with xtset ID and then include time dummies?

3) I have 20 sector dummies and that could be a problem for my regressions. Up to now I have two levels (93 company and 709 observations) and 20 sector dummies. Could be a solution that I include a third level?
1 sector-ID
2 company-ID (93 IDs)
3 observations

Thank you and best regards,
Marcus

Please create a separate topic for this, and include a data sample using -dataex-.