I wonder that will log transformation always mitigates the heteroskedasticity? Because the textbook states that log transformation often reduces the heteroskedasticity. So, I want to know that in which cases, it won't lessen heteroskedasticity.

Thanks for your reply!