xtlogit vce(cluster cvar) returning no standard errors

[Viktoria Watren]

Hello everyone,

I have a panel dataset. The dependent variable is information on whether an investment fund follows an active or passive investment strategy (1=active, 0=passive). The sample includes 36 funds and 5 years. The five independent variables are fund specific characteristics such as number of members, investment costs etc.

My advisor told me to use clustered standard errors, but Stata then returns no standard errors (for some variables) if I use the following command:
xtlogit y x1 x2 x3 x4 x5 i.year, vce(cluster cvar)

Looking at other posts in this forum, I realised that this command does not work, but I am not sure what I should use instead. I was thinking of maybe using a conditional fixed-effects model (xtlogit y x1 x2 i.a, fe) or population-averaged model with robust standard errors (xtlogit y x1 x2 i.a, pa vce(robust))

Thank you in advance for your help!

[Phil Bromiley]

Welcome to Stata list. You will increase your chances of a helpful answer by following the FAQ on asking questions-provide Stata code in code delimiters, readable Stata output, and sample data using dataex. We don't necessarily need the entire model and data – just enough to replicate the problem.

If you read the documentation for xtlogit, you'll find that is actually using the clogit procedure to do the estimates. You estimated the model with random effects. Fixed effects are unlikely to give you standard errors if random effects don't but it is trivial to check this. A population average model is very different than the fixed or random effects model – different disciplines have different preferences regarding fixed effects, random effects and population averaged estimates. As I understand it, finance tends to fixed effects. You obviously should go with what your discipline uses.

I assume cvar is your panel identifier? I'm wondering if this somehow doesn't have too many categories. You might also check if you get standard errors when you don't cluster or if you estimate the same model with xtreg.

[Jeff Wooldridge]

If you have to cluster for serial correlation then xtlogit is inconsistent. That’s why Stata didn’t allow it in earlier versions. Just use logit and cluster. Regular logit is consistent for any serial correlation pattern. Plus, joint MLE with N only 35 is pushing it.