ESTIMATING THE LINEAR-QUADRATIC INVENTORY MODEL - MAXIMUM-LIKELIHOOD VERSUS GENERALIZED-METHOD OF MOMENTS

We compare generalized method of moments (GMM) and maximum likelihood (ML) estimators of the parameters of a linear-quadratic inventory model using nondurable manufacturing data and Monte Carlo simulations. Data-based GMM estimates for five normalizations vary widely, generally rejecting the model. The ML estimate generally supports the model. Monte Carlo experiments reveal that the GMM estimates are often biased (apparently due to poor instruments), statistically insignificant, economically implausible, and dynamically unstable. The ML estimates are generally unbiased (even in misspecified models), statistically significant, economically plausible, and dynamically stable. Asymptotic standard errors for ML are 3 to 15 times smaller than for GMM.