Using nonparametric methods to improve parametric demand estimation in the presence of binding non-negativity constraints with application to agribusiness management

Jay M Lillywhite, Purdue University

Abstract

Demand analysis at an individual level, using survey data, is becoming increasingly popular. While such analysis has many benefits it suffers from computational or econometric difficulties associated with non-consumption, i.e., problems associated with binding non-negativity constraints that occur due to non-consumption of one or more goods by some survey participants. The study presented here uses a modified generalized least squares approach to estimate consumer demands from survey data. The approach consistently accounts for the role of reservation prices within the participant's consumption decision by estimating consumer reservation prices within the parameter estimation. The study examines the effectiveness of using nonparametric methods (specifically the Weak Axiom of Revealed Preference) in conjunction with parametric analysis. Two methods of incorporating nonparametric analysis are examined. First, nonparametric analysis is used to partition survey participants into sub-groups that may more meaningfully adhere to the implicit assumption that participants within the group have similar preference structures. Second, nonparametric analysis is used to construct lower bounds on reservation prices for non-consumed goods. These bounds are then integrated into the modified generalized least squares estimation procedure. Using data from a National Livestock and Meat Board Survey, the study shows that inclusion of nonparametrically derived information into parametric estimation can result in significant differences in parameter and elasticity estimates. Monte Carlo simulations are used to determine if observed differences can be considered significant improvements. The Monte Carlo simulations suggest that incorporation of nonparametric methods into parametric analysis results in not only different parameter estimates but in estimates that are more accurate. The improvements in parameter estimates were also observed in corresponding elasticity estimates. Two stylized agribusiness applications are provided in the dissertation. The first application indicates how inclusion of all information provided in a survey (including information obtained from nonparametric methods) can have significant implications for demand elasticity estimates and consequently agribusiness decisions. A second application shows how reservation price estimates obtained from the modified generalized least squares estimation may be used to target individuals or groups of individuals with special marketing and advertising programs, with the goal of increasing the profits of food manufacturers and retailers.

Degree

Ph.D.

Advisors

Preckel, Purdue University.

Subject Area

Agricultural economics

COinS