Only Time Fixed Effects for Cross-Sectional Data

#1
Hi Talkstats-Community. I have been browsing around but couldnt find an answer to my problem. Hopefully you might be able to help me on this.

I have a dataset with funds that were launched in the last 40 years and the final performance that they achieved until they were liquidated. This means I have their latest performance as of today (most were liquidated anyway, the others will be ignored), which means 1 measure per fund. I am going to analyze the performance persistence of the fund managers responsible for the funds.

From what I see this is a cross-sectional dataset (only 1 performance measurement per fund) but with funds that were started in different years. As I am only going to analyze US funds (no industry differentiation) I dont think I would need Entity fixed effects/dummies for the fundmanagers or location. I would still need Year fixed effects/dummies though because they were running over different periods of time and I would therefore like to differentiate by launch year.

My problem is:
-Am I right about this being cross-sectional data or is this an unbalanced panel?
-Would applying the year dummy make this a fixed effects model?
-As the fixed effects model seems to be usually applied to panel data it seems there is not a possibility to just run yearly fixed effects but also entity fixed effects are needed. I found something on another thread:

"The unobserved effects model is modeled as: y=Xb+u where u=c(i)+a(t)+v(it) . A one-way error model assumes a(t)=0 while a two-way error allows for a∈ℝ"

If the one-way model only considers entity effects it wouldnt be applicable as I dont need entity differentiation. Is there a way of having a one-way model that only considers Year Fixed Effects/Dummies, without applying any fixed entity effects?

Thanks a lot in advance
 

staassis

Active Member
#2
This is an unbalanced panel. You are allowed to use fixed effects for both entities and years. However, such "saturated" model is unlikely to be accurate. I would not be surprised if a model with

1) random intercept for each entity (random effect),
2) dummy variables for each year except for the reference one (fixed effect)

would be a good place to start. Many software packages have this implemented, including Stata and R (see the links).
 
Last edited: