Panel data is not possible

[Jonas Vestergaard]

I have this problem:

I am investigating which factors that effect abnormal return in football transfers. I have a clubID and year, so panel data should be perfect for this. The problem is that I have multiple observation for the same Club for each year, why I cannot use xtset ClubID Year.

Instead I can run a standard OLS and then run the regression: reg CAR1 .. i.Year i.ClubID -> but is that wrong?

Because I also know that it is possible to xtset ClubID, and then run the regression: xtreg CAR1 .. i.Year -> is that more correct than the normal OLS?

Or is there another model that is better to use? It is because I also include the person that has executed the transfer. So that person's characteristics are the same within the year for the club. This results in VERY much serial correlation, also because the firms control variable as Cash Flow, ROA and so on is included so that it will remain the same value for all the transfers for the specific club within that year. So my data is full of serial correlation. How do I cope with that, and is it possible to do?

Br. Jonas

[Carlo Lazzaro]

Jonas:
welcome to this forum.
1) in that scenario, you can simpy -xtset- your dataset with -panelvar- only. However, this workaround comes at the cost of making time-series operators (such as lags and leads) unfeasible. Obviously, if you do not plan to use them, this is not an issue. Conversely, you can still include -i.timevar- among your set of predictors.
2) you do not need to -xtset- your dataset before -regress- With your code, you tried to go two-way fixed effects (panelvar + timevar, alhough the second is not a fixed effect in a proper sense), but you forgot to cluster your standard errors on -panelvar-. By omitting this procedure, Stata interprets your dataset as composed of independent obsevations (which is not the case due to the panel structure of your dataset.
3) As an aside, plese read the FAQ on hoe to post more effectively. Thanks.

[Jonas Vestergaard]

Thank you Carlo.
I have now decided to go with panel defining only club as panel, and then include i.year variable as predictor. I get a problem when doing logit model where I have a dummy variable "Bigbuy" as dependent, if the transfer is big. Further I have the characteristics of the responsible person for executing the transfer included in the model as explanatory variables. The parameter I'm most interested to look at is how narcissistic the CEO is, and I have a measure for that which are scaled from 0-6. I want to only look at high group from 3-6 (lets call it AS), why I have created a dummy for that. Because I in the model only wants to look at them I write if AS==1. So the full model I run is:

Xtlogit Bigbuy x(all explanatory variables) i.Year if AS ==1, vce (cluster ClubID)

When I run the model I get this "convergence not achieved". I only have 73 observations. Is it possible to overcome this problem? A side note: When I do it for smallbuy it is possible, where I also only have 105 observations.

Kind regards,
Jonas

[Carlo Lazzaro]

Jonas:
this is inconsistent with your first post.
You started out reporting problems with a linear panel data regression and now that you switched to -xtlogit-, you complain about totally different issues. As you can easily figure out, this approach is not helpful.
That said, whenever a ML estimator does not converge, the usual fix is to start all over again, adding one predictor at a time and see when yout ML estimator starts gasping.
In addition, following the FAQ and posting what you typed and what Stata gave you back can help enormously those interested in your query. Thanks.