Can someone explain why conditional heteroskedasticity has consistency, but why omitted variable bias results in inconsistency and bias?
To put it very simply and without getting technical:
Conditional heteroscedasticity doesn’t violate one of the regression assumptions necessary for consistency of the estimated coefficients.
In general, OVB does violate one of the necessary assumptions for consistency and unbiasedness of the estimated coefficients (particularly, E(e|X)=0 is violated).
A technical answer would be ok.
Did my first response fail to clarify the reason?
Your comment clarified why omitted variable bias results in biased estimates. What about inconsistency?
According to the reading, multicollinearity doesn’t affect consistency and neither does serial correlation (assuming none of the independent variables are lagged values of the dependent variable).
As I mentioned in that post-- the property of consistency (and unbiasedness) can be lost because certain regression assumptions are needed for consistency-- violation of the assumption(s) will cause OLS to lose at least some of these desireable properties (this is a very basic way to explain why OLS becomes inconsistent).