Panel Regression: Direct effect of independent variables in fixed effect model

[Korhan Zeyrek]

Hey Statalists,

I am currently building a panel data set to analyse the effect of the amount of available venture capital funds on the amount of newly created firms (more precisely, the firms with a specific size). I am looking on different regions in different years, thus I need - to my current knowledge - a panel regression with state and year-fixed effects. I also have GDP as a control.

My stata line looks like this:

xtset state year
xtreg log_firm vc_funds GDP i.year, fe

If I proceed as mentioned, will I be able to see the direct effect of e.g. vc_funds on log_firm (number of firms)? I recently red that in case of state-year-fixed-effects this interpretation is flawed because the direct effect is captured by the fixed-effects? I am a bit confused here, and open for any further comments.

Your help is highly appreciated

Best
Korhan

[Carlo Lazzaro]

Korhan:
the code you posted will allow you to see the effect of -vc_funds- on the dependent variable adjusted for the remaining predictors.
I do not know the source of your last statement: -fe-specification gets rid of of time-invariant predictors and account for both observed and unobserved heterogeneity.
As an aside, if -firms- are nested within -state- you may want to consider a hierarchical random effect model (see -mixed-).

[Korhan Zeyrek]

Carlo,

Thank you very much for your insights.

I, for example, found the statement here: https://www.ecb.europa.eu/pub/pdf/sc...905d0be6a3287f (p.15). They analyzed the influence of VC funds, among others, on startups with a given size.

"Note that we are not able to identify the direct effect of VC and bank finance and industry sensitivity to VC and bank finance, as those are fully captured by the two sets of fixed effects. We thus only identify the effect of bank finance on bank-senstitive relative to bank-nonsensitigve sectors, [...]"

Not sure if this is actually comparable with the example I mentioned above. I am sorry, if not. However, it seemed to me that I am missing a point in the fixed effect regression method.

Best
Korhan

[Carlo Lazzaro]

Korhan:
as far as I can get Authorìs staterment, it seems to refer to predictors ioncluded in the right-hand side of the regression equation.
That said, the best way to share more details about your query is posting what you typed and what Stata gave you back (too) within CODE delimiters. Thanks.