Fixed effects robust gives insignificant variables

[Karel John]

Hello!

I have conducted fixed effects model using panel data. 34 OECD countries, 15 years. Total 3570 observations. Fixed effects model I already added time dummy variables to it, checked with LSDV models. Also fixed effects model is better than pooled OLS and random effects models.

When i run fixed effects model with timed fixed effects, i get that all variables are significant. Modified Wald test shows that heteroskedasticity is found.

When i use robust std. errors for fixed effects model, most variables will be insignificant.

Is that solely caused by using the robust std. errors? What can i do to make the model better?

[Carlo Lazzaro]

Karel:
welcome to this forum.
if your ran -xtreg,fe- with -robust- standard errors, you took both heteroskedasticity and autocorrelation into account.
If you have one or both those nuisances and you stick with default standard errors, your regression results will be unreliable, no matter their statistical significance.

[Karel John]

Hello Carlo!

I checked for autocorrelation using Pesaran CD test and it gave p = 0.0254, means autocorrelation is there.

To fix this, should i use first differences and run fixed effects model again to get rid of autocorrelation?

[Carlo Lazzaro]

Karel:
I'd go -xtreg,fe vce(cluster panelid)-.