When the residuals appear to be heteroscedasticity, then can we say the error terms are correlated with the independent variables ?

For me, i dont think we can say anything in terms of the correlation between indepeandent variables and the error terms, but one thing for sure is that the standard error of the estimated coefficients would be biased and estimated parameters are not bias.

Correct me if i am wrong . thanks

if it’s correlated to the independant variable then you have Conditional Heterosk.

if it’s correlated to the error t-1 then it’s Auto regressive Conditional Hetrosk

if volatility of error is not constant for no reason, then it’s unconditional Heterosk.

When L3 candidate friends and I laugh about all the stuff we had to memorize, heteroskedasticity always comes up…

thanks guys.

and which one is the bad one?

Unconditional Heteroschedasticity is usually not a problem.

Conditional H. and ARCH need to be corrected.

You can test for Conditional H using BP Chi Squared Test.

You can test for ARCH by regressing errors on their t-1 lagged values. If the slope term is statistically significant, you have ARCH (it means the squared error depends on its previous period value).

ARCH does have the benefit of allowing us to forecast variance at t+1.