The simultaneous mean-variance regression method improves prediction accuracy by considering varying dispersion in regression outcomes. It accounts for unknown heteroskedasticity and generates predictions that are better than ordinary least-squares. The approach is robust and provides valid parameter standard errors. The method dominates ordinary least-squares in the presence of heteroskedasticity and is easy to implement. It has been shown to be consistent and asymptotically normal even under model misspecification. Empirical applications have demonstrated significant improvements in estimation and inference compared to traditional methods.