This chapter examines conceptual issues concerning model prediction. Through reflection on some major ups-and-downs in the history of econometric forecasting, the defective nature is exposed of the belief that demarcates causal models from predictive models. This chapter further clarifies the conflation between causal explanation and prediction, and advocates broadening the communal understanding of prediction to embrace generalisability, namely the empirical validation of the predictive capacity of models, as part of its connotation. Acceptance of this broadened view as a prerequisite of any econometric modelling research entails a transformation of the convention of how theoretical models are produced. It also entails replacing the probability approach in econometrics with one centred around model learning following the PAC learning methodology.