On the benefits of maximum likelihood estimation for Regression and Forecasting