Statistical Cost Analysis Re-Revisited: Reply

Abstract

The issues raised by Maddala and Roberts go well beyond the question of whether pseudo-data analysis is useful empirically. It is unfortunate that their paper was not written in response to the first translog application by Christensen, Jorgenson, and Lau. This would allow separation of questions unique to pseudo-data analysis from the questions aimed more generally at the translog and other generalized functional forms. In this reply I shall attempt to separate these two questions, showing that a major portion of Maddala and Roberts' concerns relates essentially to the approximation characteristics of generalized functional forms, irrespective of the type of data used for estimation. These issues are examined in Section II. In Section III we discuss Maddala and Roberts' raising of the important question of optimal sample design and functional form, noting that there is no necessity to use orthogonal price variation and the translog function. Finally, we address in Section IV the claim that direct use of process models in place of its statistical cost function approximation is to be preferred in policy and modeling exercises.