ABSTRACT

Consistent estimation of parameters defined in a priori formulated theoretical models forms the pinnacle of Haavelmo’s probability approach. This chapter exposes the cognitive defects of this estimation-centred position. In particular, the conceptual trap of endogeneity bias and its treatment by instrumental variable estimation is dissected and highlighted. The chapter further argues that the primary function of estimation to assist the design and learning of models with meaningful and estimable/calibratable parameters of interest. It is only after models with interpretable parameters are successfully learnt that the issue of parameter inference comes onto the research agenda.