A note on aggregation bias and loss

Abstract

Since the introduction of the constant elasticity of substitution (CES) production function, estimation and testing of aggregate production functions has blossomed as never before. Concomitantly a continuing refinement of econometric techniques has enabled scholars to detect and, to some extent, correct for the potentially deleterious effects of a multitude of specification and identification problems in their work. However, the problem of aggregation bias and loss arising when an essentially micro-oriented, usually nonlinear, production model is applied to aggregate data has received scant attention. While the existence of such a problem is well known, there has been a notable lack of communication between theorists and practitioners. This is disconcerting, since the manner in which data are aggregated has a direct influence on parameter estimates, attendant conclusions and policy implications. In this note we present empirical evidence of the potential seriousness of aggregation bias and loss for a common econometric problem: estimation of parameters of an aggregate CES production function. The general tenor of our conclusions is valid for a wide range of econometric applications.