Empirical Likelihood, Maximum Entropy, Estimating Equations, and Other Computer Intensive Methods for Agricultural Economics Research

Investigators: Mittelhammer, R. C., Wahl, T.

Progress Report: Work was completed on refining methods of simulated maximum likelihood for estimation of econometric models having relatively high dimension and limited dependent variables, such as in the case of a system of consumption functions based on household survey data. A new method based on convex-hull interpolated crude frequency simulation was implemented in GAUSS computer software, applied in both Monte Carlo and empirical analyses settings, and found to have superior accuracy relative to the current standard, the GHK algorithm. A competing estimation method, based on estimating equations and the general method of moments and not maximum likelihood, was also developed, simulated, and applied in an empirical setting, and was found to be a potentially useful alternative to the simulated maximum likelihood methods. Progress continued on developing and examining the sampling behavior of a new family of estimators of econometric models consisting of single or systems of linear or nonlinear equations. Members of the family are defined via optimization of specific forms of the Cressie-Read statistic, including empirical likelihood, entropy, and log Euclidean likelihood estimation objectives. Extensive Monte Carlo simulation experiments were conducted that indicated that in problem contexts representative of empirical practice, and in small to moderate samples, the new estimators were capable of improvements in both parameter and prediction squared error loss relative to the traditional asymptotically optimal estimators such as 2SLS and optimal GMM estimators. A series of papers relating to these estimators were accepted for publication in a variety of outlets and will appear in both 2001 and 2002. Research continues in this area.Finally, refinements to a method for conducting Bayesian estimation and inference in the context of systems of equations using computationally intensive bootstrapping techniques were completed, completing a method of semi parametric Bayesian analysis free of the need to specify a likelihood function. A journal article documenting results is scheduled for publication in 2002.

Publications:

Fahs, Rafic, N.S. Cardell, and R.C. Mittelhammer. "Semiparametric Estimation and Inference In Multinomial Choice Models", selected paper presented at the summer AAEA meetings, Chicago, Illinois, August, 2001.

Hasan, Maher, R. Fahs, R.C. Mittelhammer, and T.Wahl, "Estimation of Systems of Censored Limited Dependent Variables Models using Simulated ML and GMM Estimation Methods", symposium paper presented at the summer AAEA meetings, Chicago, Illinois, 2001.

Marsh, T., R.C. Mittelhammer, and G. Judge, "Empirical Likelihood Estimators of the Linear simulataneous Equations Model", selected paper presented at the summer AAEA meetings,Chicago, Illinois, August, 2001.