Fixed effects model dummys left out due to collinearity

[Luc de Ridder]

Hello everyone,

Currently I am trying to regress a sample of approximately 1600 funds over 131 (monthly) time periods. About 180 of these are liquidated funds, 80 are merged funds and the rest are active funds. Therefore I made 3 dummys, for each of the type of funds. However, when I try to regress a fixed effects model with clustered standard errors (there is heteroskedasticity and autocorrelation and the hausman test has shown evidence for a fixed effects model) and I include the 3 dummy variables, it says the dummys are omitted due to collinearity. How can I avoid this? I would be really grateful if someone could help me.

Kind regards,

Luc

[William Lisowski]

You have a model with fund fixed effects. Your three dummies vary by fund but for a given fund do not vary over time. Thus for any given fund, the dummies are collinear with the fixed effect and therefore are omitted. The regression could make the dummy effect arbitrarily large or as small by changing the firm fixed effects in the opposite direction, so you can have firm fixed effects or dummies that are constant for a given firm, but you cannot have both.

[Luc de Ridder]

Thanks for your reply. So I have 2 options if I am correct. The first is to regress the sub samples seperately without the dummies and the second is to use a different model, but a pooled OLS is not recommended since the average value of the parameters are not constant cross-sectionally and over time. Therefore the only option is to regress the sub samples, right? Or am I missing a more appropriate model?

[Rhys Williams]

I would agree that your best options seems to run on the different sub-samples.

Perhaps a silly question - why aren't you including time fixed effects (i.e. "i.month") in your model?

Best,
Rhys