Scaling Factors in Estimation of Time-Nonseparable Utility Functions

In GMM estimations, when data exhibit exponential trends, scaling factors are often used to restore stationarity in Euler equation residuals. The present paper demonstrates that finite-sample estimates are sensitive to the scaling factors, and seemingly plausible scaling factors may produce spurious estimates. It suggests that scaling factors be chosen so that the scaled marginal utility is roughly constant. The discussion is conducted through estimation of a representative agent's time-nonseparable utility function, using first artificial data and then aggregate consumption and asset returns.