Generally it does not, as the conventional standard errors are derived assuming homoskedasticity. If your sample covers a sufficiently large number of units (e.g. firms), I'd recommend to cluster standard errors at the unit-level to account for serial correlation.